Tracking the Misallocation and Reallocation of Spatial Attention toward Auditory Stimuli

Abstract

Completely ignoring a salient distractor presented concurrently with a target is difficult, and sometimes attention is involuntarily attracted to the distractor's location (attentional capture). Employing the N2ac component as a marker of attention allocation toward sounds, in this study we investigate the spatiotemporal dynamics of auditory attention across two experiments. Human participants (male and female) performed an auditory search task, where the target was accompanied by a distractor in two-third of the trials. For a distractor more salient than the target (Experiment 1), we observe not only a distractor N2ac (indicating attentional capture) but the full chain of attentional dynamics implied by the notion of attentional capture, namely, (1) the distractor captures attention before the target is attended, (2) allocation of attention to the target is delayed by distractor presence, and (3) the target is attended after the distractor. Conversely, for a distractor less salient than the target (Experiment 2), although responses were delayed, no attentional capture was observed. Together, these findings reveal two types of spatial attentional dynamics in the auditory modality (distraction with and without attentional capture).

Significance Statement

Oftentimes, we find it hard to avoid attending to a salient sound that distracts us from our current tasks. Although a common everyday experience, little is known about how spatial distraction unfolds at the neural level in the auditory modality. Using electrophysiological markers of attention allocations, we report comprehensive evidence of spatial attentional capture by a salient auditory distractor, indicating that attention is first misallocated to the distractor and only afterward reallocated toward the target. Similar patterns were observed earlier only in vision, and their discovery in the auditory modality indicates toward the existence of domain-general spatial attentional dynamics consistent across sensory modalities. We also demonstrate that only a distractor more salient than the target reliably captures attention.

Introduction

People are constantly bombarded with a wide range of sensory experiences. To select and attend the important ones among multiple simultaneous stimuli is a massive computational challenge (Tsotsos, 1990; Bronkhorst, 2000). Likely as a side effect of these highly efficient selection mechanisms (Liesefeld et al., 2021), attention is sometimes misallocated toward a salient-but-irrelevant stimulus—leading to behavioral costs. Although almost exclusively studied in vision, this “problem” of involuntary attentional capture is also relevant for audition. Consider a typical office scenario from an auditory perspective—the sounds from mouse clicks or keyboard typing by a busy coworker, grinding of the coffee machine, footsteps of someone walking by, or even the fan noise of an overheated computer. All these sounds are perceived simultaneously and emerge from various locations—although people actively pay attention to only a small subset of them. A knock on the office door (a salient event which is relevant to the office worker) would immediately draw attention to the door. However, if a bird pecked on the window or a coworker accidentally dropped their coffee mug (an irrelevant-but-salient event), it would also capture attention.

There is little research on such spatial attentional capture by auditory stimuli, but similar scenarios have been extensively studied with visual stimuli using the additional-singleton paradigm—where a salient-but-irrelevant stimulus (distractor) occurring simultaneously with a to-be-found target stimulus involuntarily captures attention under certain conditions (Hickey et al., 2006; Kiss et al., 2012; Burra and Kerzel, 2013; Gaspar et al., 2016; Liesefeld et al., 2022). To study the attentional dynamics involved in visual attentional capture, the N2pc component of the event-related potential (ERP) has proven highly useful. The N2pc component is a transient negative increase in activity over posterior electrode sites contralateral to the attended stimulus—and it is often used as a marker of spatial attention allocations (Eimer, 1996, 2014; Constant et al., 2023). It has been used to measure the timing of attention allocations (Töllner et al., 2011; Grubert and Eimer, 2016) and, in particular, to uncover the temporal dynamics of attentional capture by salient visual distractors (Liesefeld et al., 2017, 2022; Liesefeld and Müller, 2019). Strictly speaking, the notion of attentional capture implies a series of events, involving the initial misallocation of attention to the salient distractor followed by reallocating attention toward the target. The notion of sequential attention allocations also implies that the presence of distractors delays attention allocation to the target compared with when the distractor is absent. These spatiotemporal dynamics predict a very specific N2pc pattern: (1) the occurrence of a distractor N2pc, reflecting attentional capture by the distractor, (2) a delay in target N2pc when the distractor is present compared with distractor absence, and (3) a delay in target N2pc compared with the distractor N2pc (Liesefeld and Müller, 2019).

There is evidence of an auditory analog to the N2pc which has been termed N2ac (Gamble and Luck, 2011). The N2ac component is also a transient negative potential contralateral to the location of an attended stimulus, but it is elicited by auditory stimuli and observed at more anterior electrode sites. This component is thought to be functionally similar to the N2pc component (Gamble and Woldorff, 2015; Lewald and Getzmann, 2015; Klatt et al., 2018). Thus, if it is possible to induce auditory attentional capture, we should be able to conceptually replicate the above-described pattern of spatiotemporal dynamics that has been taken as indicative of visual attentional capture. This would allow to further validate the functional interpretation of the N2ac component and to prove the existence of spatial attentional capture in the auditory domain. Most importantly, demonstrating such a relatively complex pattern of spatiotemporal attentional dynamics for the auditory modality in light of previous research in vision would be a strong evidence for more fundamental, modality-overarching principles of spatial dynamics of attentional (mis)allocation.

Experiment 1

Materials and methods

In order to build a basic auditory search scene that would allow to disentangle various attentional dynamics, we adapted the design of Hickey et al. (2009). This design provides several advantages for our purposes. Hickey et al. used a visual-search display with two objects (a square and a line), one stimulus being defined as the target while the other served as the distractor. Participants were asked to report on a feature of the target (e.g., whether the square target was a square or diamond—when the square is rotated 90°—or if the line target was a small or big line). Such tasks in which the target is defined by one feature (e.g., its shape) and participants are to report one of its features (e.g., its size) are called discrimination, categorization, or compound search tasks (Liesefeld et al., 2024) and are commonly employed in the additional-singleton paradigm (Theeuwes, 1991). Notably, by placing one stimulus on the midline and the other lateralized, Hickey et al. were able to disentangle attentional dynamics related to the respective lateralized stimulus, namely, either the target or the distractor, in the presence of the midline stimulus. Just as the N2pc studied in Hickey et al., the N2ac employed here is a component emerging contralateral to the attended stimulus, and stimuli on the midline consequently cannot elicit such lateralized activity. The additional advantage of the Hickey et al. (2009) design for our study is that only very few stimuli were presented simultaneously (also see Eimer, 1996; Hilimire et al., 2012). Presenting many identical (nontarget) stimuli simultaneously (as would be the default for visual-search studies) is not feasible for studies using spatial sounds, because the auditory system cannot spatially segregate more than a few concurrently presented sounds, in particular if these stimuli are quite abstract and similar (Bregman, 1990), like those employed in typical visual-search displays.

On this background, the auditory scene employed in the present study consisted of just two clearly distinguishable sounds (sine waves vs square-tone bursts)—where the instructions defined one as the target and the other as the distractor. Participants were required to report on each trial whether the target stimulus was of high frequency or low frequency indicated through pressing either of two keys.

Participants

Sixteen participants took part in Experiment 1 (median age, 26 years; range, 23–38 years; nine males). This sample size is in the upper range of previous N2pc and N2ac studies and sufficient to detect effects of size d_z = 0.75 and above, with a probability of 1—β = 0.80 (α = 0.05, two-tailed). In both the experiments, participants reported normal hearing. Written informed consent was obtained from all the participants, and they received course credit or were paid for their participation. No participant had to be excluded. Procedures were approved by the ethics committee of the Department of Psychology and Pedagogics at Ludwig-Maximilians-Universität Munich.

Experimental design

Stimulus presentation and response collection were controlled using PsychoPy (Peirce et al., 2019). The auditory search stimuli were presented using three studio sound monitors (Genelec 8020D, Genelec) that were placed in the frontal field (Fig. 1a). One sound monitor was placed directly in front of the participants at 0° angle, while the other two sound monitors were at ∼40° to the left and right, respectively, at 110 cm from the participants. The sound monitors were visible to participants throughout the experiment. Participants were asked to rest their head on a chinrest throughout the experiment such that there are minimal head movements and distance of the head from the sound source is kept constant.

Figure 1.

Auditory search scene. a, Setup for the auditory search. b, Representation of the waveform of the two kinds of sounds used. c, Distractor-midline/target-lateral condition. d, Distractor-lateral/target-midline condition. e, Distractor-absent/target-lateral condition. f, Both-lateral condition.

On each trial, either one (distractor-absent condition; target-only) or two (distractor-present condition) sounds were presented from different locations. The target sounds were sine-wave tones of either 440 Hz or 470 Hz. Participants’ task was to indicate whether the target was the high-frequency (i.e., the 470 Hz) or the low-frequency (the 440 Hz) tone in each trial. In distractor-present trials, an additional sound, from a different location than the target was presented. The distractor sounds were square-tone bursts (comprising of three 50 ms bursts separated by 50 ms of silence; see Fig. 1b for a visual representation of the stimuli's waveforms), which could either be of low frequency (410 Hz) or of high frequency (500 Hz). The temporal gaps in the distractor sounds render them more salient than the continuous sine-wave targets (Kayser et al., 2005). Both the distractor and the target sounds were presented for 250 ms at ∼70 dB(A) SPL, measured at the eardrum, using the miniDSP EARS device (miniDSP). Participants were instructed to maintain eye fixation at a fixation cross presented on a monitor in front of them while performing the task. The monitor was placed such that it was below the sound monitor at 0° and did not obstruct the sound from it. Participants had to respond within 3,000 ms after the onset of the sound (response deadline). In case of incorrect or delayed responses, the fixation cross changed to red or blue for 1,000 ms, respectively. No feedback was provided for correct responses. The intertrial interval was jittered between 800 and 1,200 ms.

The main part of the experiment consisted of 20 blocks of 108 trials each (2,160 trials in total), with 720 distractor-absent trials and 1,440 distractor-present trials. An additional practice block was provided at the beginning of the experiment, which consisted of 108 unanalyzed trials.

The experiments were designed to isolate attentional processing of target and distractor through lateralized ERPs—specifically the N2ac component (Gamble and Luck, 2011). The design follows a similar logic as used to dissect the attentional dynamics of target and distractor processing in visual search using the N2pc component (Liesefeld et al., 2022). The distractor-midline/target-lateral condition (Fig. 1c) contained a lateralized target with a distractor on the midline and served to isolate target-related activity. The distractor-lateral/target-midline condition (Fig. 1d) contained a lateralized distractor with a target on the midline and served to isolate distractor-related activity. The distractor-absent/target-lateral condition (Fig. 1e) contained only a lateralized target without a distractor and served as a baseline for target-related activity in the absence of a distractor. Finally, in the both-lateral condition, the distractor was presented on one and the target on the other side (Fig. 1f). Within each block, all display configurations as well as frequency of distractor and target were completely balanced and randomized in distractor-present trials, which were randomly intermixed with distractor-absent trials in which target frequency was balanced and randomized as well.

Electrophysiological recording and analysis

The EEG was recorded via 60 preamplified Ag/AgCl electrodes positioned according to the international 10–10 system. Horizontal ocular artifacts were monitored via two additional electrodes on the outer canthi of both eyes. All impedances were kept below 15 kΩ. Signals were amplified (250 Hz low-pass filter, 10 s time constant; BrainAmp DC, Brain Products) and sampled at 1,000 Hz. EEG data were processed with custom-written MATLAB scripts using functions from EEGLAB v2023.0 (Delorme and Makeig, 2004) and ERPLAB v10.0 (Lopez-Calderon and Luck, 2014) and the “latency.m” function (Liesefeld, 2018).

Signals were rereferenced off-line to the average of both mastoids. A 0.5 Hz high-pass and a 40 Hz low-pass FIR filter (EEGLAB default) were applied, after which independent component analysis was run on the signal. The independent components were classified using the ICLabel v1.4 (Pion-Tonachini et al., 2019) tool within EEGLAB, and components representing blinks or horizontal eye movements (prob. >80%) or muscle artifacts (prob. >90%) were then removed from the continuous EEG data (resulting in the removal of M = 3.94, min = 2, max = 10 components in Experiment 1 and M = 4.88, min = 1, max = 9 components in Experiment 2). After this, the data were segmented into epochs from −200 to 800 ms relative to the search-stimuli onset and baseline-corrected relative to the prestimulus interval. The trials with artifacts in the analyzed channels (FC5/6; voltage steps larger than 50 μV per sampling point, activity changes <0.5 μV within a 500 ms time window, or absolute amplitude exceeding ±80 μV; equal to M = 0.37%, min = 0% and max = 3.61% of trials in Experiment 1 and M = 2.85%, min = 0% and max = 17.23% of trials in Experiment 2) or incorrect responses were excluded (at least 367 trials remaining per individual in each condition after trial exclusion). The electrode positions FC5/6 were chosen (instead of the anterior electrode cluster used in Gamble and Luck, 2011) as the N2ac has been found to be the most prominent at these locations in previous research (Lewald and Getzmann, 2015; Lewald et al., 2016).

In order to determine the analysis window for the component of interest, the on- and offsets of the strongest component of the respective polarity were identified as the time points where the ERP in the grand-average (GA) difference wave crossed 30% of the component's peak amplitude (detected in the time window of 0–700 ms after the stimulus onset). The 50% area latency was used for the component latency estimation, where the component area was bounded by the ERP, a threshold set at 30% of the respective component's peak amplitude searched in a common time window encompassing GA on- and offsets of all analyzed components. Here and in previous work (Liesefeld et al., 2017, 2022; Liesefeld, 2018), we chose 50% area latency as our latency measure since it is more robust than other latency measures and approximates the median of the distribution of component latencies across trials (Luck, 2005; Liesefeld, 2018). Peak latency approximates a mode of the distribution, whereas percent–amplitude (onset) latency is biased toward the earliest latencies, and both earliest trials and the mode are arguably less representative of the latency distribution than the median (Rousselet and Wilcox, 2020). Both alternative measures are more prone to noise than 50% area latency, and therefore we employ a jackknife procedure (Smulders, 2010) when we calculate them for comparison. We report and interpret these alternative latency measures, i.e., jackknifed 50% amplitude (onset) latency and jackknifed peak latency, whenever at least one of their results are in conflict with results from our preferred 50% area latency.

Statistical analyses

For amplitudes, we report p_perm and p_permAdj values obtained from two permutation methods modeled after Sawaki et al. (2012). The procedure followed for p_perm is the same as that used in Liesefeld et al. (2022). For each of the 10,000 permutations, n_L and n_R trials were randomly assigned from the respective display configuration to the left and right ERP. Here, n_L and n_R represent the number of trials that went into the respective original individual ERPs after any trial rejection. The GA waveform was built from these, and the signed area amplitude (average of amplitudes in the predicted direction across the analysis window) was extracted in the time range of 0–500 ms (fixed across participants and conditions). This is 100 ms longer than that employed by Liesefeld et al. (2022; who used a 100–500 ms time window) to encompass potential earlier components which might occur due to the higher processing speed of the auditory modality (Stein and Meredith, 1993). For p_permAdj we used the same procedure as p_perm except that the analysis time range was individually adjusted to each component. In particular, comparable to the nonpermutation amplitude analysis in Liesefeld et al. (2017, 2022), we calculated the amplitude for each permutation from a 30 ms time window centered around the respective component's 50% area latency in the GA waveform in each run. Thus, this analysis time window was fixed across participants for a particular component (and run) but varied across components (and runs). The p_perm and p_permAdj values indicate the proportion of runs in the random permutation that yielded an amplitude larger than or equal to that of the correct assignment of left and right trials for the analysis. In other words, p_perm and p_permAdj can be interpreted as the probability of observing a value larger or equal to the observed amplitude merely due to random fluctuations. While robust for ERP waveforms containing a single component, p_perms Type II error rate (false negatives) might be inflated by an additional component of opposite polarity within the analysis time window (e.g., two N2pcs for stimuli on different sides in the both-lateral condition). Maybe the signed-amplitude measure of a considerable number of random permutations will be affected by the component of opposite polarity and thereby inflate the proportion of runs with a larger amplitude. Conversely, any component of the same amplitude would inflate p_perms Type I error rate (false positives), because with a broad time window, it will be counted into the signed-amplitude measure of the targeted component. Our new p_permAdj, by focusing on individualized (comparatively smaller) time windows, eliminates the need for a fixed broad time window encompassing all components that causes these problems. This permutation approach still ensures the desired Type I error rate (5%), which might even be more closely met than with p_perm in situations with multiple components of the same sign. Some of these advantages of our new method are exemplified on the present data in the analyses below.

Pairwise comparisons between distractor-absent and distractor-present conditions for median-correct reaction times (RTs) and error rates were performed using Wilcoxon signed-rank tests (two-tailed comparisons). Result graphs show the mean of medians and within-participant confidence intervals (Cousineau and O’Brien, 2014). Additionally, we report Bayes factors (BFs), for the Bayesian Wilcoxon signed-rank test (result based on data augmentation algorithm with five chains of 10,000 iterations), quantifying evidence for the alternative over the null hypothesis (BF₁₀). BFs were calculated using the standard JZS Cauchy prior with a scale factor of r = √2/2. To classify the strengths of evidence through the BFs, we used Jeffreys's criterion (van Doorn et al., 2021)—which states that for the alternative hypothesis, BFs between 1 and 3 are weak evidence, BFs between 3 and 10 are moderate evidence, and BFs greater than 10 are strong evidence. Following these criteria, 1/3 (=0.33) < BF₁₀ < 1 is considered weak evidence for the null hypothesis, 1/10 (=0.10) < BF₁₀ < 1/3 (=0.33) is considered as moderate evidence for the null hypothesis, and BF₁₀ < 1/10 (=0.10) is considered strong evidence for the null hypothesis. Accordingly, all within-experiment latency differences in the ERP components were compared using frequentist and Bayesian Wilcoxon signed-rank tests (two-tailed comparisons), while for cross-experiment comparisons, frequentist and Bayesian Mann–Whitney U tests were used. All analyses were performed using JASP v0.17.2 (JASP Team, 2023) and using custom scripts in MATLAB. The processed data, analysis scripts, and results are available on OSF (https://doi.org/10.17605/OSF.IO/QR86K).

Results

Distractor interference on behavior

Compared to distractor-absent trials (M = 526 ms; Mdn ± MAD = 516 ± 73 ms), distractor presence (M = 585 ms; Mdn ± MAD = 573 ± 87 ms) significantly delayed responses by 59 ms; W = 136.00; p < 0.001; r = 1.00; BF₁₀ = 2,582.80. The r = 1.00 means that each individual participant was delayed by distractor presence. The error rates for the distractor-present trials (M = 9.3%; Mdn ± MAD = 8.55 ± 4.45%) were also significantly higher (by 3.17%) than those for the distractor-absent trials (M = 6.13%; Mdn ± MAD = 5.94 ± 3.92%); W = 121.00; p = 0.004; r = 0.78; BF₁₀ = 32.61. Thus, overall, we observe a significant distractor interference effect in both RTs and error rates (Fig. 2a).

Figure 2.

Distractor interference in Experiments 1 and 2. a, Mean of the median RTs and error rates for the distractor-absent and distractor-present conditions in Experiment 1. b, RTs and error rates in Experiment 2. Error bars represent 95% within-participant confidence intervals. ^n.s.p > 0.05; **p < 0.01; ***p < 0.001.

Distractor presence delays attention allocation toward the target

As expected, spatial attention was (eventually) directed to the location of the target irrespective of whether the distractor was present or absent. This was evidenced by the N2ac component which was observed for the target in both the distractor-absent (p_perm < 0.001; p_permAdj < 0.001) and the distractor-midline (p_perm < 0.001; p_permAdj < 0.001) condition (Fig. 3a). Importantly, distractor presence (M = 250.63 ms; Mdn ± MAD = 250.5 ± 16 ms) delayed the target N2ac by 46.56 ms (W = 117.5; p = 0.011; r = 0.73; BF₁₀ = 5.56) compared with distractor absence (M = 204.06 ms; Mdn ± MAD = 206.5 ± 25.5 ms). This pattern of results indicates that the salient distractor indeed delayed the allocation of attention to the target. Such delay in attention allocation to the target due to a distractor is well documented in visual search using the N2pc component (Liesefeld et al., 2017, 2022). This delay is comparable with that induced by visual distractors in Liesefeld et al. (2017), 59 ms, and Liesefeld et al. (2022), 70–88 ms.

Figure 3.

Tracking the dynamics of attentional capture and filtering costs using N2ac. Top panels demonstrate the exact pattern of results as obtained from Experiments 1 (a) and 2 (b) using the N2ac component, while the bottom panels show the respective scalp topography of the contralateral minus ipsilateral waveforms at each electrode location calculated for 50 ms time windows from 0 to 700 ms. Vertical dashed lines in the ERPs (top panels) indicate mean 50% area latencies. A smoothing filter (Savitzky–Golay filter; order, 3; frame length, 51) was applied to this and subsequent waveforms before visualization to improve the visibility of effects.

Attentional capture by a salient distractor: distractor N2ac

The interference due to the distractor (as observed through the behavioral results and the delay in target N2ac due to distractor presence) could be due to the distractor entering the competition for attention and either winning this competition—thereby capturing attention or (barely) failing to win the competition and only delaying the attention toward the target—“nonspatial filtering costs” (Becker, 2007; Liesefeld et al., 2019). If attention is indeed captured by the salient distractor, then we should observe a significant N2ac toward the lateral distractor. To resolve for this, we analyzed the distractor-lateral/target-midline condition and observed a significant N2ac (p_perm = 0.025; p_permAdj < 0.001) for the distractor (Fig. 3a), thus indicating that the distractor indeed captured attention.

Attention shifts from distractor to the target

As a final criterion to consider the observed interference as attentional capture, we predicted that spatial attention as indicated by the N2ac first goes to the distractor and only afterward moves on to the target (Liesefeld and Müller, 2019). To examine this, we compared the distractor N2ac latency in the distractor-lateral/target-midline condition to the latency of the N2ac in the distractor-midline/target-lateral condition. Indeed, the N2ac toward the distractor (M = 128.19 ms; Mdn ± MAD = 135 ± 12 ms) was 122.44 ms earlier than the N2ac toward the target (M = 250.50 ms; Mdn ± MAD = 250.63 ± 16 ms); W = 136.00; p < 0.001; r = 1.00; BF₁₀ = 3,241.99 (Fig. 3a).

Another (independent) way to test for this attention shift is possible with data from the both-lateral condition (Fig. 4a; plotted and analyzed relative to the target side). There, we should observe an N2ac first toward the distractor followed by an N2ac toward the target in the same difference wave. In a difference wave calculated with respect to the target, the former should show up as a positivity, because the distractor is opposite to the target. Indeed, there was first a pronounced positivity, indicating an N2ac toward the distractor, which was followed by a negativity, N2ac toward the target (Fig. 4a). This switch in polarity of the N2ac closely resembles the N2pc polarity flip reported first by Hickey et al. (2006; but see McDonald et al., 2013) and later demonstrated more robustly by Liesefeld et al. (2017). For research on visual attentional capture, this flip has been of high theoretical significance (Theeuwes, 2010). While both N2ac components observed here were significant with our new method (p_permAdj < 0.001), only the target N2ac (p_perm < 0.001), but not the distractor N2ac (p_perm = 0.061), was significant with the more common approach.

Figure 4.

Observed and composite ERP for both-lateral conditions plotted relative to the target side for Experiment 1 (a) and Experiment 2 (b). Composite ERPs for both-lateral conditions were constructed by subtracting the distractor-lateral/target-midline condition from the distractor-midline/target-lateral condition. Top panel shows different waves, and bottom panel shows the corresponding topoplots. Interestingly, the composite ERPs seem to accentuate the patterns replicated in the actual both-lateral ERPs, which might indicate that they extract the underlying signals more accurately.

This exemplifies the advantages of our improved permutation approach when components of opposing sign occur within the same ERP: (1) from the midline conditions, we know that there is a distractor N2ac preceding the target N2ac, and (2) from the composite ERP (see the next section), we know that two components together (but with opposite sign) produce the characteristic pattern observed in the both-lateral condition. Thus, not observing a significant distractor N2ac in the latter condition must be a false negative due to a limitation of p_perm. We assume that the smaller positivity (distractor N2ac) is missed, because the analysis window also includes the target N2ac, which is larger and therefore dominates the outcome of p_perm (see Materials and Methods). By contrast, p_permAdj avoids this problem by taking a smaller time window targeted at the component of interest, which is less likely to be confounded with other components.

In any case, both sets of findings (midline vs both-lateralized conditions) clearly indicate that attention shifts from the distractor to the target. The estimated time required to shift attention from one sound to the other (122.44 ms) is within the estimates for the time required for relocating attention in visual search (100–150 ms; Woodman and Luck, 2003) and is close to the estimates empirically observed for visual targets and distractors by Liesefeld et al. (2017), 100 ms, and Liesefeld et al. (2022), 99–119 ms.

Lateralized ERPs sum up arithmetically

Presenting one stimulus in a lateral position along with another stimulus on the midline serves to isolate the attentional dynamics related to the lateral stimulus in the presence of the other stimulus. If this logic holds, the ERP for displays with the target on one and the distractor on the other side (both-lateral) should simply be the difference between the two midline conditions (Gaspar and McDonald, 2014; Liesefeld et al., 2017). Figure 4a shows this difference (composite ERP) plotted along with the actually observed both-lateral ERP. To quantify the obvious overlap between these two waveforms, we correlated both GA ERPs across the (characteristic) time window of 0–500 ms after the stimulus onset. The obtained R² = 0.80 confirms a strong overlap between the observed and composite ERP. The validity of this approach is further confirmed by the close resemblance of the scalp topography for the observed and composite ERPs (Fig. 4a, bottom panel).

Experiment 2

Materials and methods

One possible critique of the findings in Experiment 1 could be that, although completely in line with attentional capture, the N2ac pattern observed is due to low-level sensory imbalances caused by the presentation of a stimulus in one auditory hemifield and not due to attentional processes. To rule out this possibility and to also examine distractor interference without attentional capture (filtering costs), we performed a control experiment where we swap the roles of the target and the distractor from Experiment 1 (Gaspar and McDonald, 2014; Barras and Kerzel, 2017; Gaspelin et al., 2023). If the N2ac pattern observed in Experiment 1 was due to sensory imbalances in the auditory hemifield, we would expect to observe the exact same pattern in Experiment 2. If, however, the N2ac in Experiment 1 really indicates attentional capture, we should not replicate it when switching target and distractor identities. This is because the square-tone bursts employed as a distractor in Experiment 1 and as a target in Experiment 2 are much more salient than the sine-wave sounds, such that it should be much easier to select the target and to ignore the distractor (Zehetleitner et al., 2013; Gaspar and McDonald, 2014) in this experiment.

A new sample of 16 participants (median age, 24 years; range, 19–44 years; eight males) took part in Experiment 2. The general setup and the task design were similar to that of Experiment 1, except that the target- and distractor-defining features were swapped (while the reported features, i.e., the base frequencies, remained the same). That is, the target was now a square-tone burst (three 50 ms tone burst separated by 50 ms silence in between) of either 440 Hz (low frequency) or 470 Hz (high frequency). The distractors were continuous sine tones of either 410 Hz (low frequency) or 500 Hz (high frequency). Both the target and distractor were presented for 250 ms as in Experiment 1. Here again, participants were instructed to report whether the target was of high frequency or low frequency. Importantly, the switch of target- and distractor-defining features was supposed to render the target more salient than the distractor.

Results

Distractor interference on behavior

Responses on distractor-present trials (M = 529 ms; Mdn ± MAD = 500 ± 61 ms) were significantly delayed (by 14 ms; Fig. 2b) compared with distractor-absent trials (M = 515 ms; Mdn ± MAD = 497 ± 66 ms); W = 130.00; p < 0.001; r = 0.91; BF₁₀ = 362.14. Error rates showed no significant difference (Fig. 2b) between distractor-present (M = 2.42%; Mdn ± MAD = 1.32 ± 0.76%) and distractor-absent trials (M = 2.5%; Mdn ± MAD = 1.74 ± 1.04%); W = 38.00; p = 0.222; r = −0.37; BF₁₀ = 0.48.

Delay in attention allocation to the target due to distractor presence?

As in Experiment 1, here also attention was allocated to the target location—as indicated by an N2ac component that occurred when the distractor was present (p_perm < 0.001; p_permAdj < 0.001) or absent (p_perm < 0.001; p_permAdj < 0.001). According to our 50% area latency measure, there was a delay of 17.69 ms (W = 96.50; p = 0.041; r = 0.61, BF₁₀ = 3.45) for the target N2ac (Fig. 3b) when the distractor was present (M = 189.94; Mdn ± MAD = 179.0 ± 24.5) compared with when the distractor was absent (M = 172.25; Mdn ± MAD = 169.5 ± 11). This delay was significantly smaller than that observed for Experiment 1 (difference in delay between Experiments 1 and 2; M = 28.88 ms; U = 189.50; p = 0.021; r = 0.48; BF₁₀ = 1.46).

Although significant when using our preferred 50% area latency measure, this effect is not so clearly discernable from the respective ERP waveforms (Fig. 3b). Furthermore, subsequent investigation of other latency measures revealed that while jackknifed 50% amplitude (onset) latency confirms this result (p < 0.001), the jackknifed peak latency (p = 0.388) did not replicate this effect for Experiment 2. No such dependency of results on the choice of latency measure was observed in Experiment 1, for which the latency differences are also visually more evident (Fig. 3a). Thus, latency effects in Experiment 2 depend on the choice of latency measure and should be interpreted accordingly. In particular, readers not (yet) convinced that 50% area latency represents the distribution of single-trial component latencies best (see Materials and Methods) or believe that this particular latency measure is biased in some other way can still safely interpret the latency effects reported for Experiment 1, but not those for Experiment 2.

No evidence for attentional capture by the distractor: absence of distractor N2ac

Unlike in Experiment 1, where the distractor clearly captured attention as indicated by a distractor N2ac, in Experiment 2, we do not observe a statistically reliable N2ac to the distractor (p_perm = 0.306; p_permAdj = 0.577; Fig. 3b). Note that the nonsignificant p_permAdj value here demonstrates that the new method does not have a highly inflated Type 1 error probability. The absence of a distractor N2ac is further confirmed by the absence of a positivity preceding the target N2ac in the observed and composite waveform of the both-lateral condition (which again showed a strong overlap; R² = 0.80; Fig. 4b). In the absence of electrophysiological evidence for attentional capture by the less salient distractor, the distractor-presence effect on behavior (and potentially on the target N2ac latency) is best interpreted as nonspatial filtering costs (in contrast to attentional capture in Experiment 1).

N2ac latency as measure of relative salience?

As discussed earlier, the N2ac component is believed to be the auditory analog of the lateralized N2pc component (sometimes also referred to as PCN; Töllner et al., 2008). Stimulus salience is known to modulate the latency of the N2pc component elicited by a visual-search target, such that with increasing salience N2pc latency decreases (Töllner et al., 2011). To get a first impression of whether N2ac latency might also reflect relative salience, we compared distractor-absent target N2ac latencies across the two experiments since the square-tone burst (target in Experiment 2) is more salient than the sine tone (target in Experiment 1). As expected, in the distractor-absent/target-lateral condition, we indeed see a significant decrease in the target N2ac latency (Fig. 5) by 31.81 ms for Experiment 2 (more salient target) compared with that of Experiment 1 (less salient target; U = 195.00; p = 0.012; r = 0.52; BF₁₀ = 1.96).

Figure 5.

N2ac latency might indicate relative salience of stimuli. The target N2ac 50% area latency when the distractor is absent decreases for Experiment 2 compared with that for Experiment 1, indicating an increased relative salience of the target stimuli in Experiment 2 (square-tone bursts) compared with those in Experiment 1 (sine tones).

This result, however, also hinges on the validity of our preferred 50% area latency measure. Visual inspection of the respective difference waves (Fig. 5) might not be fully convincing, and other latency measures such as the jackknifed onset (p = 0.203) or peak latency (p = 0.955) do not confirm this latency difference—indicating that the earlier N2acs and the mode of the N2ac distribution are not affected by our salience manipulation. Alternatively, the failure to observe a significant effect in the alternative latency measures might be due to their lower reliability (Luck, 2005; Liesefeld, 2018). Nevertheless, in light of these discrepancies, further research, ideally employing a within-participant design along the lines of Töllner et al. (2011), is needed to substantiate the sensitivity of N2ac latency for differences in salience.

Discussion

The present study investigated the dynamics of spatial attention in auditory search using the N2ac component. In Experiment 1, an auditory distractor of higher salience than the target captured spatial attention as indicated by an N2ac to the distractor. After this initial misallocation, attention was reallocated to the target as indicated by an ensuing target N2ac, which was delayed compared with a distractor-absent condition. This signature N2ac pattern is the most comprehensive ERP evidence for attentional capture, as explained by Liesefeld and Müller (2019). In contrast, an auditory distractor less salient than the target in Experiment 2 failed to capture attention, although it still delayed responses (and maybe attention allocation toward the target), a phenomenon that is described as nonspatial filtering costs in visual search. This is the first report of both these attentional phenomena in the auditory domain.

Indication of a domain-general attention allocation mechanism

The N2pc component has been extensively studied in visual search over the last decades and has been established as a marker for the spatial allocation of attention in vision (Luck and Hillyard, 1994; Eimer, 1996, 2014; Constant et al., 2023). Similar lateralized ERP components have been discovered for other modalities, such as the N2ac component (Gamble and Luck, 2011) for audition (which we study here) and the N2cc component (Katus and Eimer, 2019; Tsai et al., 2023) for touch. Both components are believed to be functional analogs of the N2pc component in their respective sensory modalities and are used to index spatial allocation of attention.

The temporal dynamics of misallocation and consequent reallocation of spatial attention due to salient distractors has been earlier demonstrated only in vision through the N2pc component (Liesefeld et al., 2017, 2022). That the N2ac component indicates similar attentional dynamics for audition not only strengthens the functional interpretation of the N2ac component as an auditory analog to the N2pc component but also points toward the existence of domain-general spatiotemporal attentional dynamics—indicating a more fundamental principle of attentional misallocation and reallocation across all sensory modalities (see also Shinn-Cunningham, 2008 for a related discussion).

Differences in spatial attentional allocation for detection and discrimination tasks

In the original N2ac study (Gamble and Luck, 2011), the target N2ac emerged only when the target was accompanied by a distractor, whereas it emerged with and without a concurrently presented distractor in the present study (Fig. 3). This discrepancy is likely due to differences in the task designs employed in the two studies. Gamble and Luck (2011) employed a detection task, wherein participants were tasked with determining presence or absence of the target in each trial. In contrast, we employed a discrimination task, which involves the differentiation of a specific target feature (reported feature). It is well established that detection tasks impose comparatively lower attentional demands compared with discrimination tasks (Lavie, 2005) and that detection might occur in a preattentive manner (Treisman and Gelade, 1980; Luck et al., 1997). In scenarios involving target detection, an exhaustive focal–attentional engagement with the target stimuli might not be obligatory, particularly when the salience of the target is high (which is the case for the targets in Gamble and Luck, 2011). Instead, target presence can in principle be deduced from an inhomogeneity on the priority map induced by a salience signal emitted from the target (Müller et al., 2004; Liesefeld et al., 2016). This signal can inform the correct response without necessitating spatial attentional allocation and thus without eliciting an N2ac. Only when there are competing salience signals in search-detection tasks focal–attentional analysis might be required for confirming target presence (Hoffman, 1979). In contrast, discrimination of the target's reported feature (high vs low frequency in our case) in search-discrimination tasks might make a spatial shift of attention toward the target mandatory, yielding a pronounced N2ac component even in the absence of any competing stimulus.

Absence of distractor suppression (P_D) component

Along with the target N2pc component, Hickey et al. (2009) observed a P_D component—reflecting inhibition of the salient visual distractor (see Gaspelin et al., 2023 for a comprehensive review). Given the similarity of our design with Hickey et al. (2009), except for the modality difference, one could expect to observe signatures of spatial suppression in the distractor-lateral/target-midline condition. Such a P_D component (termed P_AD) has been recently reported for auditory distractors in the time range of 100–300 ms at a set of electrodes close to site FC5/6 analyzed here (Lunn et al., 2023). In this study, attention was not captured (no distractor N2ac), so that the P_D would be interpreted as proactive or stimulus-triggered suppression (Liesefeld et al., 2024) that potentially serves to avoid attentional capture (Gaspelin and Luck, 2018, 2019). When attention is captured—as was the case in our Experiment 1—such a P_D should only occur after the distractor N2ac, indicating reactive suppression (Gaspelin et al., 2023; Liesefeld et al., 2024), comparable to what has been observed in visual search (Liesefeld et al., 2017, 2022; Liesefeld and Müller, 2019).

From Figure 3, we observe some positivities contralateral to the distractor occurring in a time range of 300–600 ms in both Experiments 1 and 2. However, these positivities emerge at a time point when the target must already have been found and thus too late to be of functional significance for auditory search. Furthermore, their topography does not conform with that of the P_AD observed by Lunn et al. (2023). Rather, topoplots in Figure 3 indicate a prominent positivity over posterior electrodes in this time range in most conditions. Interestingly, a similar posterior contralateral positivity for auditory stimuli has also been observed in comparable previous studies (Gamble and Luck, 2011; Gamble and Woldorff, 2015; Lewald et al., 2016) and has been tentatively interpreted as a shift in visual attention following the auditory attentional shift, reorientation of spatial attention to a neutral position after target localization (Gamble and Luck, 2011; Gamble and Woldorff, 2015), or a posterior contralateral version of the late positive component (Lewald et al., 2016).

Differences between audition and vision

To readers familiar with the physiology of the auditory system, the observation of sound-induced lateralization might be somewhat bewildering. In contrast to the retinotopically organized visual system or the somatotopically organized haptic system, the auditory system only reconstructs the spatial source of a sound via indirect cues such as the interaural level or time differences and is generally thought to be weakly lateralized. The observed sound-induced ERP lateralization therefore indicates that attention-induced lateralization might not depend on the spatial organization of the underlying sensory system.

Gamble and Luck (2011) had argued that the lesser degree of contralaterality of the auditory system compared with the visual system might explain why the N2ac (here and in other studies) is smaller in peak amplitude (0.5–1 µV) than the typical N2pc (1–2 µV). Another reason for small peak amplitudes might be that the N2acs observed so far are less tightly locked to the sound onset and therefore smeared out across time (therefore of lower peak amplitude and longer in duration than the typical N2pc). In fact, stretched-out N2pcs of smaller peak amplitude and higher latency (i.e., smeared out N2pcs) have been observed when they likely reflect the second attention allocation (Liesefeld et al., 2017) or when the target is harder to find (Töllner et al., 2011; Dowdall et al., 2012). In the present study, the N2acs in Experiment 1 (where the target was relatively harder to find) are also relatively less in peak amplitude and longer in duration (i.e., more smeared) than those observed in Experiment 2 (where the target was relatively easier to find), thus pointing toward similar factors (in their respective modalities) influencing the amplitude and duration of both the N2ac and N2pc components.

Also note that the N2ac seems to emerge earlier than the typical N2pc. In the ERP graphs shown in Figure 3, for example, the N2ac seems to emerge (onset) already at ∼70 ms, whereas the typical N2pc does not occur before ∼130 ms [see Liesefeld et al. (2017, 2022) for examples from the same lab and Hickey et al. (2009) for search displays of comparable complexity]. Similarly, while the dynamics of N2ac and N2pc flip (in both-lateral condition) are similar, the whole complex occurs earlier in audition compared with that in vision (Hickey et al., 2006; Liesefeld et al., 2017, 2022). In line with these observations, simple reaction times to auditory stimuli are typically 20–60 ms shorter than to visual stimuli (Lewald and Guski, 2003; Shelton and Kumar, 2010) which is likely the result of faster physiological processing time in the auditory system than the visual system (Stein and Meredith, 1993). In fact, already the very first steps of perceptual processing (the transduction process) are of a magnitude faster in audition compared with that in vision (Recanzone, 2009). Thus, it is not surprising that attentional dynamics also occur earlier when triggered by auditory events compared with visual events. This would be the case even if the attentional process itself is highly comparable or even functionally identical.