Salient, Unexpected Omissions of Sounds Can Involuntarily Distract Attention Open Access

Detecting regularity violations and flexibly redirecting attention to unexpected events are important skills in everyday life (Reisenzein, Horstmann, & Schützwohl, 2019; Garrido, Sahani, & Dolan, 2013; Corbetta, Patel, & Shulman, 2008; Meyer, Reisenzein, & Schützwohl, 1997; Meyer, Niepel, Rudolph, & Schützwohl, 1991; Sokolov, 1963). For instance, a driver needs to be able to promptly notice if another driver honks at them (involuntary attention) even if they are focused on the road ahead (voluntary attention). Although it is rather well established that unexpected salient sounds (such as cell phone ringtones) can capture attention, an interesting follow-up question is whether the omission of an expected sound (e.g., a doorbell that does not ring when pressed) can also be sufficiently salient to capture attention given that, analogously to an unexpected sound, it constitutes the violation of an expectation on upcoming sensory input. This question is in line with a predictive coding account of perceptual processing.

The predictive coding theory (e.g., Friston, 2005; Rao & Ballard, 1999) proposes that, within its hierarchical cortical structure, the brain forms expectations on the upcoming stimuli based on past sensory experiences. Models of the world are built and adjusted on the basis of perceptual inference and perceptual learning: The former refers to inferring the causes of sensations, and the latter refers to learned parameters that link causes to sensory inputs. By comparing predictions coming from the current model (top–down) with the sensory inputs it receives (bottom–up), the brain attempts to minimize mismatches between the two through iterative model updates. A mismatch between prior expectations and sensations is known as prediction error, and it is thought to be reflected by multiple physiological responses (Wacongne et al., 2011; Friston, 2005).

Many studies investigated the physiological and behavioral outcomes of being exposed to an unexpected sound while engaged in an auditory or visual task that is unrelated to the sound. Typically, the experimental design is based on an oddball distraction paradigm (Escera, Alho, Winkler, & Näätänen, 1998; Schröger, 1996). In this paradigm, participants are engaged in a task (for instance, if it is a visual task, they may have to categorize pictures, or indicate whether a digit is even or odd). Before the appearance of each trial's target, they hear a sound. This sound is the same in the majority of trials (standard sound), and in rare occasions, it is a deviant or novel sound (oddball). Behaviorally, deviant or novel compared with standard sounds were shown to induce impaired performance (such as prolonged RTs or a decrease in accuracy) in the primary task not related to the sound regularity (e.g., Parmentier, Elford, Escera, Andrés, & Miguel, 2008; Andrés, Parmentier, & Escera, 2006; Escera et al., 1998; Schröger & Wolff, 1998; Alho, Escera, Díaz, Yago, & Serra, 1997; Schröger, 1996; for an early review: Escera, Alho, Schröger, & Winkler, 2000). It is commonly accepted that such behavioral outcomes can be considered distraction effects, which reflect an involuntary shift of attention away from the task-relevant stimuli toward the prediction violation.

Sounds that involuntarily distract attention because they violate predictions actually involve two prediction violations at once: the omission of an expected input and the presentation of unexpected input (Schröger, Marzecová, & SanMiguel, 2015). Moreover, when studying the processing of a mismatching sound, it is important to disentangle the responses generated by the sensory input itself (independently from predictions) and the responses to the violation of the expectation (Schröger et al., 2015). Hence, several studies have been focusing on investigating the responses to the omissions of expected sounds as they do not involve an additional prediction violation due to unexpected stimuli, and there is no potentially confounding contribution from the processing of the sensory input. How can one generate distraction through sound omissions? A prerequisite for a stimulus to be distracting is, presumably, that it violates an expectation (e.g., Parmentier, 2014). Therefore, an expectation on an upcoming stimulus needs to be created in order for its absence to be surprising. It was shown that intended outcomes of one's own actions induce very strong predictions due to action–response couplings (for a review, see Korka, Widmann, Waszak, Darriba, & Schröger, 2022). So, the omission of an intended action effect should be suited to study the distraction potential of omissions.

Using an oddball paradigm, Parmentier, Leiva, Andrés, and Maybery (2022) analyzed the effects of unexpected sound omissions on the RTs in a primary visual discrimination task in which a digit had to be classified as even or odd. At the start of each trial, there was either a sound (standard in 80% of trials or deviant in 10% of trials) or no sound (10% of trials). They found that participants had slower RTs following unexpected sound omissions compared with standard sounds.¹ Parmentier and colleagues conclude that omissions slowed RTs because, presumably, they acted as a lack of warning signal. SanMiguel, Linden, and Escera (2010) have shown that both sound-related costs (distraction caused by the deviant as an unexpected event) and benefits (facilitation caused by the deviant as an event signaling of a forthcoming target) on the processing of visual target stimuli can be additive. This may complicate the proper assessment of the processing costs of unexpected stimuli. As noted by Parmentier and colleagues (2022), although their results point to omissions having a lengthening effect on RTs by means of representing the absence of a warning signal, it remains unclear whether sound omissions additionally cause distraction because they are a violation of expected events. The present study aims at separating these two effects. The measures used are, along with RTs, pupil dilation response (PDR) and the P3a ERP as they turned out to be valuable for studying distraction.

The P3a is a positive-going ERP deflection that is elicited by regularity violating deviant sound (as compared with the ERP to the regularity-conforming standard sound). It often consists of an early part with a central distribution and a peak latency of around 250 msec (P3a or early P3a) and of a later part revealing a broader frontal and parietal distribution, peaking at around 300 msec (novelty P3 or late P3a). It is regarded as indicating a switch of attention toward the perturbating deviant sound (Polich, 2007; Escera et al., 1998; Knight & Scabini, 1998) including the enhanced evaluation of the deviating stimulus and its significance (Winkler & Schröger, 2015). It involves a distributed cerebral network with cortical generators in temporal, prefrontal, and parietal areas, but also subcortical contributions, for example, cingulate gyri and hippocampus (Alho et al., 1998; Knight, 1996; Halgren et al., 1995). The P3a has been suggested to indicate activity of the locus coeruleus (LC)-norepinephrine (NE) system elicited by motivationally significant stimuli, which—in turn—are assumed to mobilize resources for action (Nieuwenhuis, De Geus, & Aston-Jones, 2011).

P3a can be elicited when participants are exposed to a salient, for example, novel sound in a passive listening situation (Friedman, Cycowicz, & Gaeta, 2001; Escera et al., 1998), but also to an unexpected sound omission when participants intend to generate a sound, for example, via a button press (Dercksen, Widmann, Scharf, & Wetzel, 2022; Dercksen, Widmann, Schröger, & Wetzel, 2020; Korka, Schröger, & Widmann, 2019; Waszak & Herwig, 2007; Nittono, 2006). In the case of passive exposure to sounds, the prediction about the forthcoming sound is based on the extraction of the regularity inherent to the sequence of sounds. In the case of self-generated sounds, the prediction (i.e., the intended action effect) is based on the coupling between an action and a sound (Widmann & Schröger, 2022; Darriba, Hsu, Van Ommen, & Waszak, 2021; for a review, see Korka et al., 2022). It has been shown that the unexpected omission of self-generated sounds generated oN1, oN2, and oP3 ERP responses (SanMiguel, Saupe, & Schröger, 2013; SanMiguel, Widmann, Bendixen, Trujillo-Barreto, & Schröger, 2013). These were produced only when the expected sound had a specific identity associated to it as opposed to being a generic sound occurring at an expected time (SanMiguel, Saupe, & Schröger, 2013). However, Dercksen and colleagues (2020) also found (attenuated) oN1 and oP3 responses when the expected sound had a random identity. Prete, Heikoop, McGillivray, Reilly, and Trainor (2022) support previous results as they found MMN and P3a in response to unexpected sound omissions compared with expected omissions.

Similar to the P3a component, PDR has long been used as a marker of changes in attention-related states (Bradley, Miccoli, Escrig, & Lang, 2008; Stanners, Coulter, Sweet, & Murphy, 1979). Importantly, it was shown in several oddball studies that regularity-violating stimuli elicit an increased PDR when compared with standard stimuli (Selezneva, Brosch, Rathi, Vighneshvel, & Wetzel, 2021; Wetzel, Buttelmann, Schieler, & Widmann, 2016; Murphy, O'Connell, O'Sullivan, Robertson, & Balsters, 2014; Steinhauer & Hakerem, 1992; Friedman, Hakerem, Sutton, & Fleiss, 1973). Analogous results were obtained when participants were exposed to novel and emotional sounds (Bonmassar, Scharf, Widmann, & Wetzel, 2023; Bonmassar, Widmann, & Wetzel, 2020; Widmann, Schröger, & Wetzel, 2018). These modulations of the pupil's diameter reflect the neural activation patterns of the superior colliculus, whose activation is associated with the orienting responses that characterize involuntary attention, and the LC-NE system, associated with arousal and motivationally relevant stimuli (Joshi & Gold, 2020; Jepma & Nieuwenhuis, 2011; Murphy, Robertson, Balsters, & O'Connell, 2011; Aston-Jones & Cohen, 2005). A larger PDR was associated with unexpected compared with expected sound omissions (Dercksen, Widmann, & Wetzel, 2023).

The current study aimed to investigate whether distraction effects can result from the unexpected omission of a sound. However, if unexpected omissions are compared with expected sounds, a confounding factor arises: The sounds indicate the forthcoming target, that is, they serve as a warning signal speeding RTs in a primary task. This implies that prolonged RTs in the case of omissions, when compared with standard sounds, can be interpreted as either the absence of a warning signal or a result of the involuntary attentional capture induced by the occurrence of an event that violates the participant's sensory predictions. To overcome this issue, we used a paradigm partly similar to that of Parmentier and colleagues (2022) where we introduced an additional experimental condition in which sound omissions are expected (called “motor-only” or “motor” condition in the following). Furthermore, unlike Parmentier and colleagues (2022), we used an audiovisual oddball paradigm in which participants initiated each trial themselves by pressing a button. The button press generated a standard sound (80%), a deviant sound (10%), or an unexpected omission (10%). In certain blocks, the button press never generated a sound (expected omission, 100%). After this, participants were shown a visually presented digit and they had to classify it as even or odd. This design enables a dissociation of warning signal and distraction effects.

Our main hypothesis was that unexpected sound omissions would be sufficiently salient to distract attention from the primary visual task, in the same fashion as deviant sounds, as seen in prolonged RTs in the even/odd classification. Moreover, we hypothesized that the violated expectation element of the unexpected sound omissions would be associated with a P3a and greater PDR compared with expected omissions, in the same way as deviants compared with standards. The study had a 2 × 2 design with sound (sound vs. no sound/omission trials) and expectation (expected vs. unexpected sound/omission) as factors. Given our hypotheses, we expected to find a main effect of both in all of our measures (co-occurrence of warning and distraction effects). Possible results patterns additionally included the presence of interaction effects (for instance, the effect of expectation being stronger for sounds compared with no sounds, or the effect of the sound's presence or absence being stronger for unexpected events compared with expected).

Twenty-four voluntary participants from Leipzig University (20 women, 4 men, 3 left-handed, age range = 19–52 years, mean age = 23 years) took part in the experiment. We based the minimum sample size on what was indicated in the study by Parmentier and colleagues (2022). We planned to continue with sequential testing, as suggested by Schönbrodt, Wagenmakers, Zehetleitner, and Perugini (2017) in case of unclear results, but this was not necessary. All participants gave written informed consent before the experiment and received either course credits or a monetary compensation of €8 per hour for their participation. Participants reported no hearing deficits nor any history of neurological disorders, and they had either normal or corrected-to-normal vision. They were asked to not have consumed alcohol or other psychoactive substances in the 12 hr before the experiment, and they were free to stop the experimental session at any time without the need to give a reason for it. The study was approved by the ethics advisory board of Leipzig University (RF: 2022.10.21_eb_173), and it was carried out in accordance with the Declaration of Helsinki.

The experiment's auditory and visual stimuli were prepared and presented with Psychtoolbox 3 (Kleiner et al., 2007) in a Linux environment. All sounds were 100 msec long, they were tapered-cosine windowed with a 5-msec rise-and-fall and root-mean-square loudness equalized. The standard sound had a fundamental frequency of 500 Hz and included the second (−3 dB) and third harmonic (−6 dB), whereas the deviant was a pink noise sound. Every sound was presented binaurally through headphones (Sennheiser HD 25) at approximately 82 dB SPL. The visual stimuli consisted of black digits ranging from 3 to 8 presented at the center of a gray screen (10.6 cd/m²). The monitor (ViewPixx/EEG, VPixx Technologies, 120 Hz, 1920 × 1080 pixels) was placed at approximately 62 cm from the participants.

The experiment took place in a double-walled electrically shielded sound booth (Industrial Acoustics Company), where participants sat on an office chair. Each experimental session lasted a maximum of 3 hr, and it contained 11 blocks of approximately 3 min each, preceded by a practice block. Participants were free to take breaks in between blocks. In every block and for each trial, participants were presented with a fixation cross to look at. The appearance of the fixation cross was also a cue that they could initiate the trial. They initiated the trial by pressing a key with their left hand's index finger at a moment of their choice. With the key press, a standard sound played with 80% probability or a deviant sound with 10% probability or no sound played with a probability of 10%. The first and last block were half as long as the other ones, and, in these blocks, the key presses never elicited a sound (motor-only blocks, which correspond to expected omissions). The training block had 60 trials in total, the motor blocks had overall 120 trials per participant, and the remaining experimental blocks contained 120 trials each. Therefore, excluding the training, each session had a total of 1200 trials (120 expected omissions, 108 unexpected sounds, 108 unexpected omissions, 864 expected sounds). The block order and, specifically, the presence of motor-only blocks at the beginning and at the end of each session are based on an attempt to account for the participants' attentional state across the experiment as well as training effects. A similar block arrangement was also used in previous experiments (e.g., Dercksen et al., 2020; SanMiguel, Saupe, & Schröger, 2013; SanMiguel, Widmann, et al., 2013). Three hundred milliseconds after the button press, participants were presented with a digit (3, 4, 5, 6, 7, or 8) for 150 msec again followed by the fixation cross. They had to indicate whether this digit was even or odd. Depending on this, they pressed either one of two keys with their right-hand fingers (the key-digit correspondence was balanced across participants). With the discrimination key press, the fixation cross disappeared and reappeared 800 msec later, and participants were able to proceed to the next trial. Whenever the button used to start each trial was pressed before the appearance of the fixation cross, the cross appeared 800 msec after the early button press. The overall sequence can be seen in Figure 1.

For all data, the initial two trials of each experimental block were discarded, and also the two trials following a deviant or an omission and incorrect trials (those in which, e.g., the digit was odd, and the participant indicated that it was even). We additionally excluded trials in which RTs were greater than 1.25 sec or shorter than 0.15 sec. The average number of included trials in the analyses of RTs, ERPs, and PDRs per condition was 106.5 for expected omissions (95% CI [103.8, 109.1]), 391.1 for expected sounds (95% CI [380.7, 401.6])5, 102 for unexpected sounds (95% CI [99.4, 104.6]), and 101.9 for unexpected omissions (95% CI [99.5 104.4]). If a trial was excluded from either the RT, PDR, or ERPs analysis, it was excluded from all analyses (except for accuracy, which included all recorded trials). In other words, the RT, PDR, and ERP analyses are based on sets of identical trials.

EEG data were recorded with a sampling rate of 500 Hz through a 32 Ag/AgCl active electrodes system. A BrainAmp amplifier was used, and the recording was done through the Vision Recorder software (Brain Products GmbH). The setup consisted of an elastic cap (actiCAP) that followed the extended 10–20 system (Chatrian, Lettich, & Nelson, 1985) and at left and right mastoids. An additional electrode positioned on the participants' nose was used as reference, whereas a ground electrode was placed on the forehead. EOG activity was recorded through three additional electrodes positioned on the outer canthi of the eyes and under the left eye.

Data were preprocessed using the EEGLAB software (Delorme & Makeig, 2004) in MATLAB R2023a. Filtering was performed with a 0.1-Hz high-pass filter and a 48-Hz low-pass cutoff finite impulse response filter (Kaiser window, filter order 8024 for high pass and 402 for low pass; Widmann, Schröger, & Maess, 2015). Epochs were extracted from −200 to 600 msec time-locked to the button press, that is, also time-locked to sound onset (or sound omission). Noisy channels were removed whenever the robust z score of the channel's robust standard deviation exceeded a threshold of 3, except for the channels FP1, FP2, M1, M2, LO1, LO2, and IO1 (as they were required for ocular artifact component detection or could not be interpolated; see below). We removed epochs including amplitude changes exceeding a 500-μV threshold to remove large nonstereotypical artifacts (e.g., swallowing) but keep stereotypical artifacts (e.g., blinks and eye movements) for later independent component analysis (ICA) artifact removal.

Subsequently, the raw data were filtered again, this time with a 1-Hz high-pass (Kaiser window, filter order 1604) and a 48-Hz low-pass filter (same as above) to optimize artifact reduction through ICA (Klug & Gramann, 2021; Winkler, Debener, Müller, & Tangermann, 2015). They were epoched between −200 and 600 msec with respect to either the sound or the omission. The same bad channels and artifact trials as earlier were removed, and an ICA was performed with the adaptive mixture independent component analysis (AMICA) algorithm (Palmer, Kreutz-Delgado, & Makeig, 2011). The ICA unmixing matrix was copied to 0.1-Hz high-pass filtered data used for all subsequent analysis. Eye movement (presaccadic spike potentials and movements of the corneo-retinal dipoles), blink, heart, and large muscle-related ICA artifact components were identified through visual inspection by two distinct raters based on the components' topography with the aid of the ICLabel plugin (Pion-Tonachini, Kreutz-Delgado, & Makeig, 2019). Artifact activity was subtracted from the data. Then bad channels were interpolated, and the epochs were baseline corrected from −200 to 0 msec before the sound or omission. Lastly, epochs including amplitude changes exceeding a 150-μV threshold were removed.

Pupil diameter data were recorded using an EyeLink Portable Duo eye tracker (SR Research, Ottawa, Canada) at a sampling rate of 500 Hz, and the preprocessing was done in MATLAB. The pupil's diameter was converted to mm (Steinhauer, Bradley, Siegle, Roecklein, & Dix, 2022). Eye blinks detected by the eye tracker and segments with signal loss were removed from the data if they were longer than 1 sec, while shorter segments were interpolated using a piecewise cubic hermite interpolation. Partial blinks were found from the smoothed velocity time series (the function used to obtain the latter was adapted from Engbert & Kliegl, 2003) by detecting changes in pupil diameter beyond 20 mm/sec and also removed from the data or interpolated including an interval ranging from 50 msec before the blink to 100 msec after (as recommended in Merritt, Keegan, & Mercer, 1994). Given that left and right eye blinks are not perfectly simultaneous, in the segments in which a blink was (still) only present for one of the eyes, the signal segments from the other eye was interpolated (Kret & Sjak-Shie, 2019). To be considered, the signal between two blinks was required to have a minimal temporal length of 10 msec or a maximal temporal distance of 50 msec to the next adjacent valid data segment. Data were epoched from 0.2 sec before to 1.8 sec following the button press, and they were baseline-corrected by subtracting the mean pupil diameter of a 0.2-sec prestimulus (or omission) onset window. Finally, epochs were rejected if the whole baseline or more than half of the whole epoch was interpolated.

Statistical results for the ERPs were computed for the following canonical ROI: Cz, FC1, FC2, C3, C4, CP1, CP2. There were two peaks of interest (early and late P3a peaking 250 and 310 msec relative to event onset), and the statistics were computed on the mean amplitudes in time windows of a 40-msec duration built around the respective peaks.

For the statistical analysis of the PDR, we computed the mean pupil diameter in a 1.028- to 1.428-sec time window (centered on the peak of the average of the grand-average PDR responses across conditions) per participant and stimulus type.

Data were analyzed with Bayesian repeated-measures ANOVA (see van den Bergh et al., 2020; Rouder, Morey, Verhagen, Swagman, & Wagenmakers, 2017) based on a 2 × 2 within-subject design with the factors Expectation² (expected vs. unexpected) and Sound (sound vs. omission). The conventions used to interpret the results are those suggested by Lee and Wagenmakers (2014), according to which BF₁₀ > 3 indicates moderate evidence, BF₁₀ > 10 strong evidence in favor of the alternative hypothesis, and 0.3 < BF₁₀ < 3 is considered anecdotal evidence, whereas BF₁₀ < 0.33 is an indication for a moderate evidence and BF₁₀ < 0.1 is considered a strong evidence for the null hypothesis.

The ANOVAs were computed through the function anovaBF from the R package BayesFactor (Version 0.9.12.4.4; Morey et al., 2022), the priors used were r = .5 for the fixed effects and r = 1 for random effects (the default “medium” effect size priors), and all Bayes factors were calculated with 250,000 sample repetitions (higher than the default value of 10,000 to ensure the results' robustness). All the alternative models were tested against the null model. BF_incl values were found across matched models to evaluate the contribution of the effect by comparing all the averaged models including versus excluding the effect. They were computed through the function bayesfactor_inclusion from the package bayestestR (Version 0.13.1; Makowski, Ben-Shachar, & Lüdecke, 2019).

Two-sided, paired Bayesian t tests (ttestBF function from the BayesFactor package) were used in case an interaction effect emerged from the ANOVA. A noninformative Jeffreys prior was placed on the normal population's variance, and a Cauchy prior (with the default scaling factor of r = .707) was placed on the standardized effect size (Morey et al., 2022; Rouder, Morey, Speckman, & Province, 2012; Jeffreys, 1961).

The analyses were conducted with R 4.3.1 (R Core Team, 2014).

The means including 95% CIs of all dependent variables are reported in Table 1. Figure 5 shows the bar plots of the dependent variables, and the violin plots of the differences between the unexpected and expected conditions.

Accuracy was higher following unexpected events whereas the presence or absence of a sound did not affect behavioral accuracy. The Bayesian ANOVA favored the model that included only the Expectation main effect (BF₁₀ = 1.82). Accuracy was slightly higher on unexpected trials than on expected trials. The data provide only anecdotal evidence for the main effect of Expectation (BF_incl = 1.81). Note that unlike the statistics for RTs, PDR, and ERP, which exclude the first two trials per block and two events following unexpected trials as well as incorrectly responded and trials affected with PDR or ERP artifacts, the statistics for accuracy include all trials.

RTs were faster for sound compared with no-sound trials and also faster for expected versus unexpected sounds or omissions. The Bayesian ANOVA preferred the model including both main effects of Sound and Expectation (BF₁₀ = 139) but not their interaction. The data provided strong evidence for the Sound (BF_incl = 14.9) and Expectation main effects (BF_incl = 13.7) and moderate evidence against a Sound × Expectation interaction effect (BF_incl = 0.294).

PDR was larger for sound compared with no-sound trials, and it was also larger for unexpected versus expected sounds or omissions. Similar to the case of RT, the Bayesian ANOVA indicated that the best model included both main effects of Sound and Expectation but not their interaction (BF₁₀ = 3653.67; but note that the effects of Expectation on RTs and PDRs were synergistic—unexpected events caused longer RTs and higher PDRs—but the effects of Sound were antagonistic—sound events caused shorter RTs but higher PDRs). The data provided strong evidence for the Sound (BF_incl = 81.78) and Expectation main effects (BF_incl = 109.54) and moderate evidence against a Sound × Expectation interaction effect (BF_incl = 0.315). The PDR trends for all conditions can be seen in Figure 4B.

The ERP data were analyzed for both the early and the late components of the P3a. The ERP waveforms can be seen in Figure 2. Figure 3 shows the difference between the unexpected and expected waveforms with and without sound (for the sound condition, it is standard vs. deviant; in the no-sound condition, it is expected vs. unexpected omission) and the corresponding topographic potential maps (Figure 3B).

An early P3a (more positive ERPs in unexpected compared with expected sounds or omissions) was observed in both sound and no-sound trials, but it was larger in sound trials. Here, the best model included not only a main effect of both Sound and Expectation but also their interaction (BF₁₀ = 1.14 * 10¹⁶). Bayesian t tests showed strong evidence for an effect of Expectation for sound trials (standards vs. deviants: BF₁₀ = 1.18 * 10⁶) and anecdotal evidence for an effect of Expectation for no-sound trials (motor-only vs. unexpected omission: BF₁₀ = 2.64).

A late P3a was found in both sound and no-sound trials. The best model is analogous to that obtained for the early P3a. More specifically, it includes a main effect of Sound, Expectation, and their interaction (BF₁₀ = 5.01 * 10¹²). Here, Bayesian t tests showed strong evidence for the alternative hypothesis (there is a significant difference in the late P3a between unexpected and expected) in both the sound condition (BF₁₀ = 1.98 * 10⁵) and the no-sound condition (BF₁₀ = 151.44).

We presented participants with an audiovisual oddball paradigm. Every trial was initiated by the participant by button press. Together with the button press, a standard sound was presented in most trials (80%; expected sound), a deviant sound in some trials (unexpected sound; 10%), and no sound in some trials (unexpected omission; 10%). No sounds were presented in two additional motor-only blocks (expected omissions; 100%). That is, our experimental design had four conditions: unexpected omissions, expected omissions, unexpected sounds, and expected sounds. This allowed us to separately analyze the outcomes of the presence versus absence of a sound (warning effect), and those of the occurrence of an unexpected versus expected event (distraction effect). Following these auditory events, a visual digit was presented and participants were asked to indicate whether the digit was even or odd as fast and accurately as possible. Behaviorally, we observed that RTs were faster when there was a sound compared with when there was no sound (indicating a sound-induced facilitation), whereas unexpected events slowed down RTs (indicating that unexpected events caused distraction). In pupillometry, we see that the presence of a warning signal induced greater pupil dilation than no sound, as well as unexpected events. In the ERPs, there was strong evidence for P3a in response to unexpected events enhanced in sound compared with omission trials, corresponding with the presence of a warning signal.

Auditory distraction is thought to originate through a range of different potential mechanisms. Two important contrasting accounts are novelty detection (e.g., Cowan, 2008; Gati & Ben-Shakhar, 1990; Sokolov, 1963) and expectancy violation (e.g., Vachon, Hughes, & Jones, 2012; Winkler, Denham, & Nelken, 2009; Bendixen, Roeber, & Schröger, 2007; Hughes, Vachon, & Jones, 2007; Näätänen, 1990). According to the expectancy violation view, distraction is caused by the perception of an event that is not coherent with what should be expected from the pattern presented up to that point. In the novelty detection view, instead, the lack of a recent memory representation (referred to as neural model in Sokolov, 1963) associated with the stimulus is what elicits distraction. However, it was shown that auditory attentional capture can also be elicited by stimuli already encoded in the recent auditory memory if they are not expected in a specific context, which seems to favor the expectancy violation view (e.g., Marois, Pozzi, & Vachon, 2020; Parmentier, 2014; Vachon et al., 2012; Hughes et al., 2007; Näätänen, 1990). Our results support this view as well, given that unexpected omissions induced prolonged RTs in the visual task when compared with expected omissions, even though they did not represent a new sound as they had no sound at all.

Parmentier and colleagues (2022) compared the behavioral outcomes of rare deviants and rare omissions to those of standard sounds, respectively, and they found prolonged RTs in the primary task for omissions, leading the authors to conclude that rare omissions are processed primarily as the removal of a warning signal. Our results confirm that the absence of a sound elicits prolonged RTs, presumably because of the absence of a warning. However, they also show that rare omissions are processed as unexpected events, which capture attention, resulting in prolonged RTs. The warning and distraction effects on RTs are independent and additive. It is interesting to note that a warning signal effect could be observed despite the fact that (a) it was predictable and (b) the participants already had full information about the occurrence of the target, as they started each trial themselves. Thus, the warning signal could not add task-relevant information, but presumably improved performance by increasing arousal (see below for discussion).

The trials of this study were always initiated by the participants themselves. After the fixation cross appeared on the screen in front of them, they were free to start the trial whenever they were ready. The fact that the button press predictably (and immediately with a latency fo 2 msec) elicited a sound in most trials yields strong predictive power, as participants had exact information on the upcoming stimuli. Despite this, participants had still, presumably, enhanced arousal when exposed to acoustic warning signals as suggested by the reduced RTs as well as the increase in PDR corresponding to sounds as opposed to no sounds. It is possible that a sound needs to be salient enough to act as an effective warning signal to elicit a distraction response when, in rare instances, it is missing.

In conclusion, our RTs' results replicate various audiovisual oddball studies because RTs were prolonged following unexpected sounds compared with expected. However, they were also longer for unexpected omissions compared with expected, which shows that prediction is also important for determining the effect of omissions on performance, not only the fact that a warning signal is missing.

Our pupillometry results are consistent with results from other studies on unexpected sound omissions. Dercksen and colleagues (2023) compared the PDR induced by unexpected and expected omissions to analyze the pupillary response to surprise, and they found stronger pupil dilation when the auditory omission was unexpected. In addition, they found an analogous effect for somatosensory omissions as well, suggesting that pupil dilation as a measure of surprise is not modality specific. Our results are also similar to those of, for instance, Wetzel and colleagues (2016) and Bonmassar and colleagues (2023) in relation to deviant sounds, in that unexpected events induce a greater PDR compared with expected events.

It is important to point out that the PDR may constitute, almost perfectly overlapping, the alerting response to sound stimuli, the increased arousal in response to unexpected stimuli, and the response to the target digit onset (reflecting both pupil dilation due to expected task demands as well as luminance changes). Regarding the latter, this was present in all trials and therefore its addition of noise to the pupillometric signal was not considered a source of concern. In response to unexpected sounds, we additionally observed an early deflection of the PDR, which was also observed in the biphasic pattern reported previously by Steinhauer and Hakerem (1992) or Wetzel and colleagues (2016). This component might reflect sphincter muscle-related parasympathetic inhibition, and it has almost no overlap with the later and larger deflection, which presumably reflects dilator muscle-related sympathetic activation.

Some studies investigated directly attentional capture (seen behaviorally or through changes in the PDR) induced by novel stimuli, as compared with deviants that stay the same throughout the experiment (e.g., Marois et al., 2020; Hughes et al., 2007). In Marois and colleagues (2020), two types of spoken letter sequences were presented to participants while they were engaged in a visual task: a type of sequence in which one of the letters was novel, and a type of sequence in which one letter had already been presented before but was in an unexpected position. Both these types of deviant events impaired task performance equally and generated a PDR, but the PDR's amplitude was more pronounced for novel deviants. This showed that, while it significantly affects it, novelty is not necessary to elicit a PDR (Marois et al., 2020). This is confirmed by our results, given that there is no acoustic novelty in unexpected omissions, and yet they elicit a PDR significantly larger than that of expected omissions.

Notably, both warning and distraction effects were found because both sounds and unexpected events had an effect on the PDR (specifically, they induced a larger PDR). In previous research, auditory stimuli were shown to elicit a PDR (for a review on how different eye responses including pupil dilation are modulated by sounds, see Hu & Vetter, 2024), which was enhanced by a greater saliency and loudness of the sounds (Liao, Kidani, Yoneya, Kashino, & Furukawa, 2016). The warning effect of sounds on the PDR found in the current study could derive from the arousal induced by the presence of sounds as opposed to no sounds.

A clear antagonistic pattern between RT and PDR emerges in the results: Warning signals (expected as well as unexpected sounds) shorten RTs and lead to enhanced PDRs. Distraction of attention by unexpected events (unexpected sounds and omission) prolonged RTs and lead to enhanced PDRs. Thus, warning signals and distraction by unexpected events produce opposing effects on the two dependent variables. This antagonistic result pattern proves that we could successfully dissociate the two effects by means of our experimental paradigm and therefore provide evidence for distraction of attention by salient omissions.

The presence of sounds, as opposed to their absence, has a warning effect on participants presumably because of the enhancement of arousal it can induce. As a consequence, participants were faster and their pupils dilated when exposed to sounds. Instead, the violation of an expectation induces an orienting response, reflected by the results: While an orienting response translates to a behavioral distraction effect (the prolonged RTs), it also induces an increased dilation in the pupil as also seen in recent literature on PDRs to infrequent sound omissions (Dercksen et al., 2023).

For both RT and PDR, we found evidence against a Sound × Expectation interaction effect. The presumed lack of interaction highlights the independence of these variables from each other, and it is coherent with them reflecting different processes (sounds enhance arousal, unexpected events induce an orienting response).

These patterns show a clear distinction of warning signal and distraction, because although both cause the phasic dilation of the pupil due to LC-NE activation, they have opposite effects on the RTs (warning signals speed up RTs, while violation-induced distraction prolongs them). Crucially, rare omissions induced prolonged RTs, which suggests that they can act as distractors.

We observed more positive ERP amplitudes in response to unexpected events in both sound and omission trials. For sound trials, the two peaks in the unexpected minus expected difference wave can be clearly identified as early and late P3a by latency and topography. For omission trials, no clearly distinguishable peaks could be identified (possibly due to latency jitter due to the less well-defined event onset), but the functional sensitivity and the observed topographies in the corresponding analysis time windows were at least well compatible with an interpretation as early and late P3a. We discuss alternative interpretations below.

The P3a ERP component, associated with distraction, has been extensively studied in the case of distraction induced by deviant or novel sounds (for a review, see Polich, 2007). More recently, it was additionally observed in relation to the omission of expected sounds. Our findings are in line with the literature because we found P3a elicited in response to unexpected events compared with expected both sound and no-sound events. More specifically, we found strong evidence for an effect of expectation in the sound condition for both early and late P3a, whereas for omissions, the evidence was weak for the early P3a and strong for the late P3a. Lastly, we found a strong interaction effect of sound and expectation.

Masson and Bidet-Caulet (2019) found that what in the current study is referred to as early P3a is strongly related to the arousal value of distracting sounds. However, the early P3a may also reflect continued or enhanced stimulus processing (Horváth, Sussman, Winkler, & Schröger, 2011). The weakness of the evidence for an effect of expectation on the early P3a in the case of omissions, together with the strength of the evidence for sounds, suggests the possibility that unexpected events that are not accompanied by sensory signals are either less arousing and/or their processing is not enhanced (as they do not contain new sensory information).

Our late P3a, instead, could be closely linked to attentional capture or orienting of attention (Masson & Bidet-Caulet, 2019; Bidet-Caulet, Bottemanne, Fonteneau, Giard, & Bertrand, 2015; Escera et al., 1998, 2000). On the basis of this, the strong effect of expectation for both sounds and omissions supports that both events acted as salient distractors.

The early and late P3a observed in response to unexpected sound omissions have significantly smaller amplitudes and do not show as clearly distinguishable peaks in the ERP difference waves as the early and late P3a in response to unexpected sounds. Although the interpretation as early and late P3a is plausible to our understanding, the positive deflection could also be (partly) due to differences in amplitude or latency of other components. For example, for expected events (sounds and omissions), we observed a negative trend over central and parietal electrode locations, which can be interpreted as a contingent negative variation (CNV) ERP component (Brunia, van Boxtel, & Böcker, 2012; Tecce, 1972; Walter, Cooper, Aldridge, McCallum, & Winter, 1964) driven by the contingency between the button press (and possibly the expected event) and the delayed imperative and task-relevant visual stimulus, reflecting preparatory brain activity. The observed positive deflection in unexpected compared with expected sound omission ERPs should then be interpreted as absence (or modulation) of CNV in response to unexpected omissions. Note, however, that in any case, we would categorize the absence of preparatory activity in response to an unexpected stimulus as a deviance- or surprise-related modulation of processing and thus as a distraction effect, and consequently an observed modulation of CNV (or the modulation of the amplitude or latency of other overlapping ERP components) as an electrophysiological signature of distraction. This supports our conclusion that the slower RTs following unexpected sound omissions presumably reflect behavioral distraction, which is based on our main findings, the absence of an Expectation × Sound interaction effect on RTs and PDRs, and is further supported by the antagonistic effects of sound and expectation on RTs and PDRs.

The LC is an ensemble of noradrenergic neurons in the brainstem whose main transmitter is NE (also called noradrenaline; for reviews, see Poe et al., 2020; Sara & Bouret, 2012). The LC-NE system has been long associated with arousal, which is closely related to attention and motivation (Aston-Jones & Cohen, 2005), and its activation was linked to both the P3a and the PDR in several human and nonhuman primate studies (e.g., Joshi, Li, Kalwani, & Gold, 2016; Murphy et al., 2011, 2014; Pineda, Foote, & Neville, 1989). Moreover, it was hypothesized that P3a and PDR activations may find a shared source in a region in the medulla that projects to both the LC-NE system and the peripheral sympathetic nervous system (Nieuwenhuis et al., 2011).

According to our results, PDR and the amplitudes of both subcomponents of the P3a were larger for sounds compared with no sounds, and unexpected events were followed by a P3a and larger PDR. The similar trends of P3a and PDR could be attributed to the involvement of the LC-NE system due to the motivational significance of both sounds and prediction violations.

The motor control blocks were presented at the beginning and end of the experiment to balance potential effects of arousal and attentional state, and did not include unexpected events to keep the experiment as short as possible. Both were done similarly in previous studies (e.g., Dercksen et al., 2020; SanMiguel, Widmann, et al., 2013). Arousal and attentional state may not have been perfectly balanced, as participants may have been more aroused in the first block or more fatigued in the last block. The level of expected uncertainty (see, e.g., Yu & Dayan, 2005) might have been different in the motor-control blocks compared with the experimental blocks, which also included rare unexpected events (deviant sounds and sound omissions). Note, however, that the sounds or omissions were not relevant to the task, participants were explicitly instructed to ignore them, and two trials at the beginning of the block and after each unexpected event were excluded. We found no evidence for asymmetric effects and did not observe a Sound × Expectation interaction effect on RTs and PDR (rather evidence against such an interaction) that would be predicted by differences in expected uncertainty between block types. Thus, we assume that our conclusions are not confounded by the fixed block order and the absence of unexpected events in the motor control blocks.

The current study's main contribution is that it successfully teases apart the two processes of the brain reacting to a warning signal that directs attention to the task at hand, and of the distraction orienting attention away from the task toward a salient, unexpected auditory event. Our results show electrophysiological, pupillometric, and behavioral correlates of the two processes, suggesting that they are both present and independent for RTs and PDRs. Interestingly, we observed an effect of attention in the RTs and in the pupil dilation that has an antagonistic nature: Although warning signals are responsible for an enhancement of attention directed toward task-relevant stimuli, which can be observed through lower RTs and more dilated pupils, events that violate prior expectations distract from the task, causing higher RTs and yet again more pupil dilation.

The saliency of omissions as distractors might be confined to predictions based on action–effect couplings, which are known to facilitate the establishment of predictions (Waszak & Herwig, 2007; Nittono, 2006; for a review, see Korka et al., 2022). Ultimately, our results support previous findings that involuntary attentional mechanisms are grounded in the violation of sensory predictions (Schröger et al., 2015; Parmentier, 2014). We provide evidence that unexpected sound omissions, and not only deviant sounds, can be sufficiently salient to distract from a task at hand, despite being effectively the absence of a stimulus.

The authors would like to thank Benjamin Eichenberger and Annika Löhr for their support in data collection.

Corresponding author: Valeria Baragona, Wilhelm Wundt Institute for Psychology, Leipzig University, Neumarkt 9-19, Leipzig, 04109, Germany, e-mail: [email protected].

The averaged data and R scripts used for the statistical analysis are available here: https://osf.io/ymdzb/. The raw data are available from the authors upon request.

Valeria Baragona: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Software; Visualization; Writing—Original draft; Writing—Review & editing. Erich Schröger: Conceptualization; Funding acquisition; Methodology; Project administration; Resources; Supervision; Writing—Review & editing. Andreas Widmann: Conceptualization; Data curation; Formal analysis; Methodology; Software; Supervision; Validation; Visualization; Writing—Review & editing.

This study was funded by the Bundesministerium für Bildung und Forschung (https://dx.doi.org/10.13039/501100002347), grant number: M526300.

Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance. The authors of this paper report its proportions of citations by gender category to be: M/M = .56; W/M = .214; M/W = .155; W/W = .071.