Intact sensory processing but hampered conflict detection when stimulus input is task-irrelevant

Conflict detection in sensory input is central to adaptive human behavior. Perhaps unsurprisingly, past research has shown that conflict may be detected even in the absence of conflict awareness, suggesting that conflict detection is a fully automatic process that does not require attention. Across six behavioral tasks, we manipulated task relevance and response overlap of potentially conflicting stimulus features to test the possibility of conflict processing in the absence of attention. Multivariate analyses on human electroencephalographic data revealed that neural signatures of conflict are only present when at least one feature of a conflicting stimulus is attended, regardless of whether that feature is part of the conflict. In contrast, neural signatures of basic sensory processes are present even when a stimulus is completely unattended. These data reveal an attentional bottleneck at the level of objects, suggesting that object-based attention is a prerequisite for cognitive control operations involved in conflict detection.


35
Every day we are bombarded with sensory information from the environment and we often face the challenge 36 of selecting the relevant information and ignoring irrelevant -potentially conflicting-information to maximize 37 performance. These selection processes require much effort and our full attention, sometimes rendering us 38 deceptively oblivious to irrelevant sensory input (e.g. chest-banging apes), as illustrated by the famous 39 inattentional blindness phenomenon (Simons & Chabris, 1999). However, unattended events that are not 40 relevant for the current task might still capture our attention or interfere with ongoing task performance, for 41 example when they are inherently relevant to us (e.g. our own name). This is illustrated by another famous 42 psychological phenomenon: the cocktail party effect (Cherry, 1953;Moray, 1959). Thus, under specific 43 circumstances, task-irrelevant information may capture attentional resources and be subsequently processed 48 VanRullen, 2007). It was long thought that only basic physical stimulus features or very salient stimuli are 49 processed in the absence of attention (Treisman & Gelade, 1980), due to an "attentional bottleneck" at higher 50 levels of analysis (Broadbent, 1958;Deutsch & Deutsch, 1963

60
Over the years, various theories have been proposed with regard to this attentional bottleneck among which 61 the load theory of selective attention and cognitive control (Lavie et al., 2004), the multiple resources theory 62 (Wickens, 2002), the hierarchical central executive bottleneck theory and formalizations thereof in a cortical 63 network model for serial and parallel processing (Sigman & Dehaene, 2006;Zylberberg et al., 2010Zylberberg et al., , 2011. These 64 theories all hinge on the idea that resources for the processing of information are limited and that the brain 65 therefore has to allocate resources to processes that are currently most relevant, via selective attention 66 (Broadbent, 1958;Treisman, 1969). Resource (re-)allocation, and thus flexible behavior, is thought to be 67 governed by an executive network, most prominently involving the prefrontal cortex (Goldman-Rakic, 1995, 68 1996. Information that is deemed task-irrelevant has fewer resources to its disposal and is therefore likely 69 processed to a lesser extent. When more resources are necessary for processing the task-relevant information, 70 e.g. under high perceptual load, processing of task-irrelevant information diminishes (Lavie et al., 2003(Lavie et al., , 2004.

71
Yet even under high perceptual load, task-irrelevant features can be processed through object-based attention 72 (Chen,

80
Specifically, we aim to test whether cognitive control operations, necessary to identify and resolve conflicting 81 sensory input, are operational when that input is irrelevant for the task at hand (and hence unattended).

82
Additionally, we test whether in the absence of task-relevance of the conflicting feature, object-based attention 83 4/39 might overcome any hampered control operations. Previous work has shown that the brain has dedicated 84 networks for the detection and resolution of conflict, in which the medial frontal cortex (MFC) plays a pivotal 85 role (Ridderinkhof et al., 2004). Conflict detection and subsequent behavioral adaptation is central to human 86 cognitive control and, hence, it maybe not surprising that past research has shown that conflict detection can 87 even occur unconsciously (Atas et

91
Conclusive evidence regarding the latter claim has, to our knowledge, not been provided and therefore the 92 necessity of attention for cognitive control operations remains open for debate. Previous studies have shown 93 that cognitive control processes are operational when to-be-ignored features from either a task-relevant or a 94 task-irrelevant stimulus overlap with the behavioral response to be made to the primary task, causing 95 interference in performance (Mao & Wang, 2008;Padrão et al., 2015;Zimmer et al., 2010). In these 96 circumstances the interfering stimulus feature carries information related to the primary task and are therefore 97 de facto not task-irrelevant. Consequently, it is currently unknown whether cognitive control operations are 98 active for conflicting sensory input that is not related to the task at hand. Given the immense stream of sensory 99 input we encounter in our daily lives, conflict between two (unattended) sources of perceptual information is 100 inevitable.

102
Here, we investigated whether conflict between two features of an auditory stimulus (its content and its spatial 103 location) would be detected by the brain under varying levels of task-relevance of these features. The main 104 aspect of the task was as follows. We presented auditory spoken words ("left" and "right" in Dutch) through 105 speakers located on the left and right side of the body. By presenting these stimuli through either the left or the 106 right speaker, content-location conflict arises on specific trials (e.g. the word "left" from the right speaker) but 107 not on others (e.g. the word "right" from the right speaker) (

113
There were several critical differences between the behavioral tasks: (1) task relevance of a conflicting feature,

114
(2) task relevance of a non-conflicting feature that was part of a conflicting stimulus and (3) whether the 115 response mapped onto a conflicting feature (e.g. respond right when left/right were conflicting features). Note 116 that in all tasks, only one feature could be task-relevant, and that all the other feature(s) had to be ignored. The 117 systematic manipulation of task-relevance and the response mapping allowed us to explore the full landscape 118 of possibilities of how varying levels of attention affect sensory and conflict processing. Electroencephalography 119 (EEG) was recorded and multivariate analyses on the EEG data was used to extract any neural signatures of 5/39 while being presented with the same auditory stimuli-all features of which being fully irrelevant for task 133 performance. Behavioral responses on this visual task were orthogonal to the response tendencies potentially 134 triggered by the auditory features, excluding any task or response related interference (see figure 1B). Under 135 this manipulation, all auditory features are task-irrelevant and are orthogonal to the response-mapping. To 136 maximize the possibility of observing conflict detection when conflicting features are task-irrelevant, and to 137 explore the effect of task automatization on conflict processing, participants performed the tasks both before 138 and after extensive training, which may increase the efficiency of cognitive control (see figure 1C) (

141
To investigate whether our experimental design was apt to induce conflict effects for task-relevant sensory input 142 and to test whether conflict effects were still present when sensory input was task-irrelevant, we performed 143 repeated measures (rm-)ANOVAs (2x2x2 factorial) on mean reaction times (RTs) and error rates (ERs) gathered 144 during the EEG-recording sessions (session 1, "before training"; session 4, "after training"). This allowed us to 145 include 1) task-relevance (yes/no), 2) training (before/after) and 3) congruency of auditory content with location 146 of auditory source (congruent/incongruent). Note that congruency is always defined based on the relationship 147 between two features of the auditorily presented stimuli, also when participants performed the visual task (and 148 therefore the auditory features were task-irrelevant). In both tasks, the spoken words "left" and "right were presented through either a speaker 152 located on the left or right side of the participant. Important to note is that auditory stimuli are only task-relevant in auditory content 153 discrimination task I and not in the vertical RDM task. In this figure, sounds are only depicted as originating from the right, whereas in the 154 experiment the sounds could also originate from the left speaker. (A) In content discrimination task I, participants were instructed to report 155 the content ("left" or "right") of an auditory stimulus via a button press with their left or right hand, respectively, and to ignore the spatial 156 location at which the auditory stimulus was presented. (B) During the vertical RDM task, participants were instructed to report the overall 157 movement of dots (up or down) via a button press with their right hand, whilst still being presented with the auditory stimuli, which were 158 therefore task-irrelevant. In both tasks, content of the auditory stimuli could be congruent or incongruent with its location of presentation 159 (50% congruent/incongruent trials). (C) Overview of the sequence of the four experimental sessions of this study. Participants performed 160 two EEG sessions during which they first performed the vertical RDM task followed by auditory content discrimination task I. Each session 161 consisted of 1200 trials, divided over twelve blocks, allowing participants to rest in between blocks. In between experimental sessions, 162 participants were trained on auditory content discrimination task I on two training sessions of 1 hour each.

207
Therefore, we adopted an additional hypothesis-driven analysis, which also allowed us to obtain evidence for 208 the absence of effects. Throughout this paper, we will discuss our neural data in the following order.   figure S2). Therefore, we subsequently conducted the decoding 234 analysis on merged data from before and after the extensive behavioral training in the conflict-inducing content 235 discrimination task I, thereby maximizing power to establish effects in our crucial comparisons. Thus, results 236 reported here were obtained through the multivariate analyses of EEG-data of the two sessions combined.

261
Similarly, the location of auditory stimuli could also be decoded from neural data for both content discrimination 262 task I (p<0.001, one-sided: x̅ >0.5, cluster-corrected; frequency-range: 2Hz-24Hz, peak frequency: 6Hz, time- The experimental design of the first experiment rendered the auditory features to be located at the extreme 291 ends of the scale of task-relevance, i.e. either the conflicting features were task-relevant and the conflicting 292 features were consistently mapped to specific responses, or the conflicting features were task-irrelevant and 293 the conflicting features were not mapped to responses. However, to further understand the relationship 294 between the relevance of the conflicting features and the overlap with responses, we performed a second 295 experiment containing four behavioral tasks. For this second experiment we recruited 24 new participants. We 296 included two auditory conflicting tasks, similar to content discrimination task I. In one of the auditory tasks (from 297 hereon: content discrimination task II, figure 3A) participants again had to respond according to the content of 298 the auditory stimulus, whereas in the other auditory task (from hereon: location discrimination task, figure 3B) 299 they were instructed to report from which side the auditory stimulus was presented (i.e. left or right speaker). 300 Furthermore, we included two new tasks in which the conflicting features (location and content) were 301 not task-relevant and participants responded to a non-conflicting feature that was part of the conflicting 302 stimulus (from hereon: volume oddball detection task, figure 3C) or the auditory stimulus was task-irrelevant 303 but its features location and content overlapped with the responses to be given (from hereon: horizontal RDM 304 task, figure 3D). The horizontal RDM task was similar to vertical RDM task of experiment 1, however the dots 305 were now moving in a horizontal plane. Thus, participants were instructed to classify the overall movement of 306 moving dots to either the left or the right. As this is a visual paradigm, the simultaneously presented auditory 307 stimuli are fully task-irrelevant. However, both features of conflict, the content (i.e. 'left' and 'right') and the 308 location (i.e. left and right speaker), of the auditory stimuli could potentially interfere with participants' 309 responses on the visual task, thereby inducing a crossmodal Stroop-like type of conflict (Stroop, 1935).

310
In the volume oddball detection task, participants were presented with the same auditory stimuli as 311 before, however occasionally (1/8 trials) they were presented at a lower volume. Participants were instructed 312 to detect these volume oddballs by pressing the spacebar with their right hand as fast as possible. If they did not 313 hear an oddball, they were instructed to withhold from responding. Participants performed 500 trials of the 314 volume oddball detection task twice, at the very beginning of a session and at the end of a session (figure 3E).

315
The rationale behind this was to see if effects in both runs differed because in the intermediate tasks sound 316 location and sound content were relevant. In other words, during the first run sound content and sound location 317 had never been related to any behavioral responses, whereas during the second run they might have acquired 318 some intrinsic relevance through training on the other tasks, despite the fact that they were not task-relevant 319 within the context of the task the participant is performing at that time. Theoretically, the selection of an object's 320 feature (e.g. volume) may lead to the selection of all of its features (e.g. sound content, location), as suggested 321 by theories of object-based attention (Chen, 2012), which may lead to conflict detection if this would be the 322 case.

323
Again, EEG was recorded while participants performed these tasks, in order to see if auditory conflict 324 was detected when the auditory stimulus, or its conflicting features (i.e. location and content) were task-325 irrelevant. In order to keep sensory input similar, moving dots (coherence: 0) were presented on the monitor 326 during content discrimination task II, the location discrimination task and the volume oddball detection task, but 327 these could be ignored. discrimination task II, participants were instructed to report the content ("left" or "right") of an auditory stimulus via a button press with 334 their left or right hand, respectively, and to ignore the location of the auditory stimulus that was presented. (B) In the auditory location 335 discrimination task, participants were instructed to report the location (left or right speaker) of an auditory stimulus via a button press with 336 their left or right hand, respectively, and to ignore the content of the auditory stimulus that was presented. (C) During the volume oddball 337 task, participants were instructed to detect auditory stimuli that were presented at a lower volume than ordinary stimuli (i.e. oddballs) by 338 pressing the spacebar with their right hand. (D). In the horizontal RDM, participants were instructed to report the overall movement of dots 339 (left or right) via a button press with their left and right hands respectively, whilst still being presented with the auditory stimuli. In all four 340 tasks, content of the auditory stimuli could be congruent or incongruent with its location of presentation (50% congruent/incongruent trials).

341
(E) Order of behavioral tasks in experiment 2. Participants always started with the volume oddball task, followed by the location 342 discrimination task, content discrimination task and horizontal RDM in randomized order and ended with another run of the volume oddball 343 task.

345
We will first discuss the behavioral results of the content discrimination, location discrimination and horizontal 346 RDM tasks, as behavioral data for the volume oddball task is represented in d', rather than ER. Repeated 347 measures ANOVAs (3x2 factorial) were performed on mean RTs and ERs from these three tasks, with the factors 348 1) task and 2) congruency of the auditory features. Again, congruency always relates to the combination of the 349 auditory stimulus features sound content ("left" vs. "right") and sound location (left speaker vs. right speaker).

351
We observed that participants were slower and made more errors on incongruent trials as compared to

376
Experiment 2: Detection of conflict occurs when any feature of a stimulus is task-relevant (even a 377 non-conflicting feature), but never for a stimulus that is entirely task-irrelevant

378
We again trained multivariate classifiers on single-trial time-frequency data to test whether the auditory 379 stimulus features (i.e. content, location and congruency) were processed when (I) the auditory conflicting 380 features were task-relevant and overlapped with the response-mapping (content and location discrimination 381 tasks), (II) the auditory conflicting features were task-irrelevant and another stimulus of the conflicting stimulus 382 was task-relevant (volume oddball task), or when (III) the auditory conflicting features was task-irrelevant, but 383 its conflicting features overlapped with the response-mapping in the task (horizontal RDM task) (see Methods).

385
Neural signatures of conflict processing in the theta-band were clearly present when one of the conflicting 386 features was task-relevant, i.e. both when the content of the auditory stimulus was task-relevant (content 387 discrimination task II: p<0.001, one-sided: x̅ >0.5, cluster-corrected; frequency-range: 2Hz-20Hz, peak frequency: 388 4Hz, time-range: 266ms-906ms, peak time: 438ms; figure 4B) and when the location of the auditory stimulus 389 was task-relevant (location discrimination task: p=0.002, one-sided: x̅ >0.5, cluster-corrected; frequency-range: 390 2Hz-6Hz, peak frequency: 4Hz, time-range: 328ms-813ms, peak time: 516ms; figure 4B). However, for both the 391 volume oddball task and the horizontal RDM task, no significant clusters of above-chance classifier accuracy 392 were found after multiple comparisons ( figure 4B). This was surprising for the volume oddball task, because 393 even though a non-conflicting feature was task-relevant, strong and significant behavioral measures of conflict 394 detection were observed in this task (e.g. d' higher for incongruent than congruent trials, see figure 4A). Again, 395 we followed up these hypothesis-free (with respect to frequency and time) MVPA analysis with a hypothesis 396 driven analysis focused on the post-stimulus theta-band (2Hz-8 Hz, 100ms-700ms). Indeed, as expected, this 397 more restricted analysis revealed significant above chance decoding in the volume oddball task for congruency 398 (t23=2.29, p=0.016, d=0.47, BF10=3.66), in agreement with the behavioral results. Although significant, this 399 conflict effect was clearly weaker than for the tasks where one of the conflicting features was task-relevant.

400
When no feature of the conflict-containing stimulus was task-relevant, no effects of conflict were apparent 401 (t23=0.00, p=0.49, d=0.00, BF01=4.65). Therefore, to qualify the differences between tasks, below we combine 402 the data from all experiments and compare effect sizes across tasks (see accuracy were only present for the two auditory tasks where both the auditory stimulus was attended and at least one of its conflicting 417 features was task-relevant, namely content discrimination task II and the location discrimination task. Information related to auditory 418 congruency was observed in theta-to alpha band clusters for these tasks. For the volume oddball task and the horizontal RDM no above-419 chance decoding clusters were observed. Note that the data shown for the volume oddball task was merged over both sessions.  4) and an alpha/beta-band cluster (p<0.001, cluster-corrected; frequency-range: 12Hz-24Hz, 435 peak frequency: 18Hz, time-range: 219ms-891ms, peak time: 766ms; figure 4B).

437
Furthermore, sound location could also be decoded from all tasks: content discrimination II (p<0.001, one-sided: 438 x̅ >0.5, cluster-corrected; frequency-range: 2Hz-14Hz, peak frequency: 4Hz, time-range: 203ms-750ms, peak performed a 6x3 rm-ANOVA on these accuracies, with the factors task and stimulus feature. We found main 474 effects for behavioral task and stimulus feature (task : F 1,2.17 =20.27, p<0.001  In two omnibus experiments with 6 different tasks, we presented potentially auditory-spatial conflicting 511 stimulus features (e.g. the word "left" presented on the right side) to participants, whilst they were performing 512 several behavioral tasks. These tasks manipulated whether the features sound content and location of the 513 auditory stimulus were task-relevant and whether these features were mapped to specific overlapping 514 responses to the primary task. We observed clear signals of conflict processing in behavior (i.e. longer reactions 515 times, increased error rates, increased sensitivity) and brain activity (i.e. above chance decoding accuracy in the 516 theta-band) when the conflicting features of the auditory stimulus were task-relevant, i.e. in the content and 517 location discrimination tasks, and when another non-conflicting feature of the auditory stimulus was task-518 relevant, but the conflicting features content and location were not (volume oddball task). When none of the 519 features of the auditory stimulus were task-relevant, i.e. in the vertical/horizontal RDM task, we did not observe processed by the brain (see figure 5A). Interestingly, conflict between two task-irrelevant features was only 540 detected when another feature of the conflicting stimulus was the task-relevant stimulus (volume oddball task), 541 but absent when none of the auditory features were task-relevant (RDM tasks). We argue that this difference is 542 due to the fact that in the volume oddball task, the task-irrelevant conflicting features were selected through conflict was decodable when the auditory conflicting features were task-relevant coincides with the time-range 561 in which these features could be decoded when the auditory conflicting features were task-irrelevant (see figure   562 5A). Therefore, it seems unlikely that the more temporally constrained processing of task-irrelevant stimulus 563 features is the cause of hampered conflict detection.

565
Besides time being a factor, the processing of task-irrelevant features in the RDM tasks may have also been too 566 constrained to (early) sensory cortices and therefore could not progress to integration networks, including the 567 MFC, necessary for the detection of conflict. Speculatively, the processing of task-irrelevant auditory features 568 was relatively superficial, due to the relatively few remaining resources (

588
First, we explicitly separate task-relevant stimulus features that cause conflict and task-relevant features that 589 do not, parsing the cognitive components that induce cognitive control in this context. Furthermore, in the RDM 590 and volume oddball tasks we tested whether conflict between two task-irrelevant features could be detected 591 by the brain. Specifically, we investigated if conflict between two task-irrelevant features would be detected in 592 the presence or absence of object-based attention (volume oddball task vs. RDM tasks), also manipulating 593 whether task-irrelevant conflicting features mapped onto the response (horizontal RDM) or not (vertical RDM).

594
This approach is crucially different from previous studies that exclusively tested whether a task-irrelevant or 595 unattended stimulus (feature) could interfere with processing of a task-relevant feature (Mao & Wang, 2008; 596 Padrão et al., 2015;Zimmer et al., 2010). Under such conditions, at least one source contributing to the 597 generation of conflict (i.e. the task-relevant stimulus) is fully attended and therefore one cannot claim that under 598 those circumstance conflict detection occurs outside the scope of attention.

600
It can be argued that in our horizontal RDM task, the task-irrelevant auditory features (location and content) 601 that mapped onto the response of the primary task could interfere with the processing of horizontal dot-motion, 602 i.e. the task-relevant feature. This is in fact true, as we found effects of auditory content-dot motion and auditory 603 location-dot motion conflict in RT and ER (see supplementary figure S6). This nicely highlights that a single 604 feature of a task-irrelevant stimulus can interfere with the response to a task-relevant stimulus when there are 605 overlapping feature-response mappings. This is different from two features of a task-irrelevant stimulus to 606 produce inherent conflict (e.g. between auditory content and location), which is what we specifically 607 17/39 investigated by always testing the presence of auditory content-location conflict only. A similar argument might 608 be made for our vertical RDM and volume oddball tasks, because in those cases the auditorily presented stimuli 609 could potentially conflict with responses that were exclusively made with the right hand, e.g. the spoken word 610 "left" or the sound from left location may conflict generally more with a right hand response (independent of 611 the up/down classification or oddball detection) than the spoken word "right" or the sound from right location.

612
In the vertical RDM task, where the auditorily presented stimuli were truly task-irrelevant, as both stimulus 613 content and location did not affect mean RTs (see supplementary figure S7). In the volume oddball task sound 614 content and location were task-irrelevant features, but these features were subject to object-based attention 615 as they were part of the attended stimulus. In this behavioral task, the content of the auditory stimuli interfered 616 with responses to the volume oddball task, resulting in longer RTs (see supplementary figure S8). Moreover, 617 under this manipulation we did find behavioral and neural effects of conflict between two auditory features (see 618 figure 4). The absence of conflict effects in the horizontal RDM and presence of such effects in the volume 619 oddball task strengthens our conclusion that at least one feature of the stimulus containing the conflicting 620 features should be task-relevant -and thus the presence of object-based attention-is what ultimately 621 determines whether conflict can be detected. Summarizing, we show that the brain is not able to detect conflict 622 that emerges between two features of a task-irrelevant stimulus (in the absence of object-based attention).

624
Lastly, in other studies conflicting stimuli were often task-irrelevant on one trial (e.g. because they were 625 presented at an unattended location) but task-relevant on the next (e.g. because they were presented at the 626 attended location) (e.g. Padrão et al., 2015;Zimmer et al., 2010). Such trial-by-trial fluctuations of task-relevance 627 allow for across-trial modulations to confound any current trial effects (e.g. conflict-adaptation effect) and also 628 induce a "stand-by attentional mode" where participants never truly disengage to be able to determine if a 629 stimulus is task-relevant. We prevented such confounding effects in the present study, where the (potentially) 630 conflicting features or the auditory stimulus itself were task-irrelevant on every single trial in the vertical RDM, 631 horizontal RDM and volume oddball task.

633
Differences between response conflict and perceptual conflict cannot account for absence 634 of conflict detection in task-irrelevant sensory input 635 One difference between the content and location discrimination tasks on the one hand and the volume oddball 636 and RDM tasks on the other, was the task-relevance of the (conflicting) auditory features. Another major 637 difference between these groups of tasks was, consequently, the origin of the conflict. When the auditory stimuli 638 were task-relevant, the origin of conflict was found in the interference of a task-irrelevant feature on behavioral 639 performance, whereas for the other tasks this was not the case. We argued that in the volume oddball and RDM 640 tasks, salient auditory stimuli could be intrinsically conflicting. Intrinsic conflict is often referred to as perceptual 641 conflict, as opposed to the aforementioned behavioral conflict (Kornblum, 1994). Although perceptual conflict 642 effects are usually weaker than response conflict effects, both in behavior and electrophysiology (Frühholz et

657
Inattentional deafness or genuine processing of stimulus features? 658 The lack of conflict effects in behavior and EEG when auditory conflict features were task-irrelevant (in the 659 absence of object-based attention) might suggest a case of inattentional deafness, a phenomenon known to be 660 induced by demanding visual tasks, which manifests itself in weakened early (~100ms) auditory evoked 661 responses (Molloy et al., 2015). Interestingly, human speech seems to escape such load modulations and is still tasks. For all tasks, classification of stimulus content was most accurate in the theta-band which is in line with a 669 previously proposed theoretical role for theta-oscillations in speech processing, namely that they track the 670 acoustic-envelope of speech (Giraud & Poeppel, 2012). In the volume oddball detection task, we also observed 671 an alpha/beta-band cluster of above-chance decoding (ranging from 12-24Hz, peak in 18Hz), which has 672 previously been linked to single word intelligibility (Becker et al., 2013). After this initial processing stage, further 673 processing of stimulus content is reflected in more durable broadband activity for the content discrimination 674 tasks, possibly related to higher-order processes (e.g. semantic) and response preparation (figure 4, two left 675 columns).

677
Similarly to processing of stimulus content, early processing of stimulus location was most strongly reflected in 678 the delta-to theta-range for all tasks (figure 4), which may relate to the auditory N1 ERP component, an ERP

700
In order to see whether enhancing such automaticity could hypothetically increase the likelihood of conflict 701 detection, we included extensive training sessions in the first experiment and measurements of the volume 702 oddball task before and after exposure to conflicting tasks in the second experiment. In the first experiment, we 703 found no neural effects of conflict detection in the vertical RDM task, even when participants had been trained 704 19/39 on the auditory task for 3600 trials (see supplementary figure S2). Training did result in decreased conflict 705 effects in the auditory content-discrimination task of experiment 1, indicating that our training procedure was 706 successful. Further, participants improved within-trial conflict resolution (see supplementary figure S1), 707 suggesting more efficient functioning of conflict resolution mechanisms. In experiment 2, participants 708 performed the volume oddball task twice, once before and once after sound location and content had been 709 mapped to responses. Again, we aimed to see if training on conflict tasks would enhance automaticity of conflict 710 processing in a paradigm where the auditory conflicting features were task-irrelevant. We did not find any 711 statistically reliable differences in behavioral conflict effects (d', RT, see supplementary figure S9) or accuracy 712 of congruency decoding between the two runs (see supplementary figure S10). Therefore, it seems that the 713 automaticity of conflict detection by the MFC and associated networks do not hold when the auditory stimulus 714 is task-irrelevant (at least after the extent of training and exposure as studied here). In the task-relevant auditory conflict task, the spoken words "links" (i.e. "left" in Dutch) and "rechts" (i.e. "right" 756 in Dutch) were presented through speakers located on both sides of the participant ( figure 1A). Auditory stimuli 757 were matched in duration and sample rate (44 kHz) and were recorded by the same male voice. By presenting 758 these stimuli through either the left or the right speaker, content-location conflict arose on specific trials (e.g. 759 the word "left" through the right speaker). Trials were classified accordingly, as either congruent (i.e. location 760 and content are the same) or incongruent (i.e. location and content are different). Participants were instructed 761 to respond as fast and accurate as possible, by pressing left ("a") or right ("l") on a keyboard located in front of 762 the participants, according to the stimulus content, ignoring stimulus location. Responses had to be made with 763 the left or right index finger, respectively. The task was divided in twelve blocks of 100 trials each, allowing 764 participants to rest in between blocks. After stimulus presentation, participants had a 2s period in which they 765 could respond. A variable inter-trial-interval between 850ms-1250ms was initiated directly after the response.

766
If no response was made, the subsequent trial would start after the 2s response period. Congruent and 767 incongruent trials occurred equally often (i.e. 50% of all trials), as expectancy of conflict has been shown to affect 768 conflict processing (Soutschek et al., 2015). Due to an error in the script, there was a disbalance in the amount

776
Participants were instructed to respond according to the direction of the dots, by pressing the "up" or "down" 777 key on a keyboard with their right hand as fast and accurate as possible. Again, participants could respond in a 778 2s time-interval which was terminated after responses and followed by an inter-trial-interval of 850-1250ms.

779
Task difficulty, in terms of dot-motion coherency (i.e. proportion of dots moving in the same direction) was 780 titrated between blocks to 73-77% correct of all trials within that block. Similar to content discrimination task I, 781 the vertical RDM was divided in twelve blocks containing 100 trials each, separated by short breaks. Again, 782 congruent and incongruent trials, with respect to the auditory stimuli, occurred equally often.

784
Experiment 2: design and procedures 785 In the second experiment, we wanted to investigate whether it is task-irrelevance of the auditory stimulus itself 786 or task-irrelevance of the auditory features (i.e. content and location) that determine whether prefrontal control 787 processes are hampered. Participants performed two tasks in which auditory stimuli were fully task-relevant 788 (location-discrimination and content-discrimination), one task in which the auditory stimulus was relevant but 789 the features auditory location and content were not (volume oddball) and one task in which the auditory 790 stimulus itself was task-irrelevant, but its features location and content could potentially interfere with behavior 791 (horizontal RDM). Participants came to the lab for one session, lasting three hours. Each session started with the 792 volume oddball task, followed by (in a counterbalanced order) the location discrimination, content 793 discrimination and horizontal RDM tasks, and ended with a block of the volume oddball task again.

794
We included the location discrimination and content discrimination tasks both to replicate the results of The auditory content discrimination task from experiment 2 is a (near identical) replication of the auditory 803 content discrimination task from experiment 1. Participants were fixating on a fixation mark in the center of the 804 screen. Again, the spoken words "links" (i.e. "left" in Dutch) and "rechts" (i.e. "right" in Dutch) were presented 805 through speakers located on both sides of the participant. Participants were instructed to respond according to 806 the stimulus content by pressing left ("A") or right ("L") on a keyboard located in front of them, with their left 807 and right index fingers respectively. Concurrently, on every trial, a black disk with randomly moving dots 808 (coherence: 0) was presented to keep sensory input similar between tasks. After stimulus presentation, 809 participants had an 800ms period in which they could respond. After a response, the response window would 810 be terminated directly. A variable inter-trial-interval between 250ms-450ms was initiated directly after the 811 response. If no response was made, the subsequent trial would start after the ITI. All stimulus features (i.e. 812 sound content, location and congruency) were presented in a balanced manner (e.g. 50% congruent, 50% 813 incongruent). The task was divided in six blocks of 100 trials each, allowing participants to rest in between blocks.

814
Auditory location discrimination task 815 The auditory location discrimination task was identical to the auditory content discrimination task II, with the 816 exception that participants were now instructed to respond according to the location of the auditory stimulus.

22/39
Thus, participants had to press a left button ("A") for sounds coming from a left speaker and right button ("L") 818 for sounds coming from a right speaker. Again, participants performed six blocks of 100 trials. 819 820 Volume oddball task 821 In the volume oddball task, the same auditory stimuli were presented. Again, on every trial, a black disk with 822 randomly moving dots (coherence: 0) was presented to keep sensory input similar between tasks. Occasionally 823 an auditory stimulus would be presented at a lower volume. The initial volume of the oddballs was set to 70% , 824 but was staircased in between blocks to yield 83%-87% correct answers. If participants performance on the 825 previous block was below or above this range the volume increased or decreased with 5% respectively. The odds 826 of a trial being an oddball trial were 1/8 (drawn from a uniform distribution). Participants were instructed to 827 detect these oddballs, by pressing the space-bar as fast as possible whenever they thought they heard a volume 828 oddball. If they thought that the stimulus was presented at a normal volume, they were instructed to refrain 829 from responding. The response interval was 800ms, which was terminated at response. A variable inter-trial-

847
Data analysis 848 We were primarily interested in the effects of congruency of the auditory stimuli on both behavioral and neural 849 data. Therefore, we defined trial-congruency on the basis of these auditory stimuli, in all behavioral tasks of the 850 two experiments. All behavioral analyses were programmed in MATLAB (R2017b, The MathWorks, Inc.).

852
Analysis of behavioral data 853 Missed trials and trials with an RT <100ms or >1500ms were excluded from behavioral analyses. In order to 854 investigate whether current trial conflict effects were present both when conflict was task-relevant and task-855 irrelevant and to inspect if training of the auditory task affected such conflict effects, we performed rm-ANOVAs 856 on RTs and d'-scores for the volume oddball task of experiment 2 and on RTs and ERs for all other tasks. For the 857 tasks from experiment 1, we performed these ANOVAs with task-relevance, training (before vs. after) and 858 current trial congruency as factors (2x2x2 factorial design). Additional post-hoc rm-ANOVAs, for content 859 discrimination task I and vertical RDM task separately (2x2 factorial design), were used to inspect the origin of 860 significant factors that were modulated by task-relevance. If the assumption of sphericity was violated, we 861 applied a Greenhouse-Geisser correction.

862
For content discrimination task I, the location discrimination task and horizontal RDM task we performed a rm-863 23/39 ANOVA with task and congruency as factors (3x2 factorial design). Additional post-hoc paired sample t-tests 864 were performed per task to test the difference between incongruent and congruent trials.

865
In case of null findings, we performed a Bayesian analysis (rm-ANOVA or paired sample t-test) with identical 866 parameters and settings on the same data, to test if there was actual support of the null hypothesis (JASP Team, 867 2018).  Epochs were created by taking data from -1s to 2s around onset of stimulus presentation.

881
Multivariate pattern analysis (decoding) 882 We applied a multivariate backwards decoding model to EEG data that was transformed to the time-frequency 883 domain. We used multivariate analyses both because its higher sensitivity in comparison with univariate 884 analyses and to inspect if and to what extent different stimulus features (i.e. location and content) were 885 processed in both tasks, without having to preselect spatial or time-frequency ROIs. The ADAM-toolbox was 886 used on epoched EEG data, that were transformed to time-frequency using default methods of the toolbox but 887 with specific settings (epochs: -100ms to 1000ms, 2Hz-30Hz in linear steps of 2Hz) (Fahrenfort et al., 2018). Trials 888 were classified according to current trial stimulus features (i.e. location and content) resulting in 4 trial types.

889
As decoding algorithms are known to be time-consuming, epochs were resampled to 64Hz. Next, a decoding 890 algorithm, using either stimulus location, stimulus content or congruency as stimulus class, was applied on the 891 data according to a tenfold cross-validation scheme. Specifically, a linear discriminant analysis (LDA) was used 892 to discriminate between stimulus classes (e.g. left versus right speaker location etc.). Classification accuracy was

895
The multivariate classifiers were on different subsets of trials, depending on the behavioral task. For the auditory 896 tasks (content discrimination I and II, location discrimination and volume oddball detection), only correct trials 897 were included in the analysis as errors tend to elicit a similar, albeit not identical, neural response as cognitive 898 conflict. Errors are more likely on incongruent trials and therefore error trials had to be excluded. For the volume 899 oddball detection task we additionally excluded all oddball trials, thus only testing correct rejections, in order to 900 prevent conflict arising between responses made exclusively with the right hand and sound content and 901 location. For the two visual tasks (horizontal and vertical RDM) we trained the classifier on all trials.

903
The AUC scores we obtained via multivariate analyses of our EEG data were tested per time-point and frequency 904 with one-sided t-tests (x̅ >0.5) across participants against chance-level (50%). These t-test were corrected for 905 multiple comparisons over time, using cluster-based permutation tests (p<0.05, 1000 iterations). This procedure 906 24/39 yields time clusters of significant above-chance classifier accuracy, indicative of information processing. Note 907 that this procedure yields results that should be interpreted as fixed effects (Allefeld et al., 2016), but is 908 nonetheless standard in the scientific community.

909
In addition to the cluster analysis, we performed a hypothesis-driven analyses on classifier accuracies that were 910 extracted from a pre-defined time-frequency ROI. Specifically, for every task and every stimulus feature (i.e. 911 congruency, content, location), we extracted mean ROI accuracies per subject. This ROI was selected visually, to 912 cover a time-and frequency-range in which most significant clusters could be found. The ROI ranged from 913 100ms-700ms and 2Hz-8Hz. We then applied a 6x3 rm-ANOVA on these accuracies with factors task and stimulus 914 feature. Next, one-sample t-tests (one-sided, x̅ >0.5) were performed on every task/feature combination to 915 determine whether decoding accuracy of a specific feature within our pre-selected ROI was above chance during 916 the various behavioral tasks. Additional Bayesian one-sample t-tests (one-sided, x̅ >0.5, Cauchy scale=0.71) were 917 performed to inspect evidence in favor of the null-hypothesis that decoding accuracy was not above chance.

1196
Thresholded (cluster-base corrected, p<0.05) accuracies are depicted across the frequency-range (2-30Hz) and significant clusters are 1197 outlined with a solid black line. For both content discrimination task I and vertical RDM, classifier accuracy was not affected by extensive 1198 training of the task.

1199
Excluding task-related effects of content discrimination task I on decoding in vertical RDM 1200 In our experimental design, we excluded across-trial fluctuations of task-relevance of the auditory stimulus, by 1201 applying a blocked design. This ensured that the auditory stimulus in the vertical RDM task was task-irrelevant 1202 on every single trial. However, participants were extensively trained in the conflict-inducing content 1203 discrimination task I, and as such participants might have been prompted to detect the auditory stimuli in the 1204 second session of the vertical RDM task, due to their earlier relevance. This in turn might have confounded our 1205 decoding results, for which we merged data from both EEG sessions, before and after training in content 1206 discrimination task I. That is, the above chance decoding of stimulus content in the vertical RDM task might 1207 mainly be caused by some sort of lingering background process aiming to detect the previously task-relevant 1208 stimulus content of the auditory stimulus. However, stimulus content could already be decoded from data of 1209 the first session of the vertical RDM task, when participants were not yet exposed to content discrimination task