Figure 4. Behavioral and multivariate decoding results for experiment 2.

(A, B) The four columns show data belonging to, from left to right, content discrimination task II, the location discrimination task, the volume oddball detection task, and the horizontal random dot-motion (RDM) task. (A) Behavioral results are plotted as conflict effects (incongruent – congruent). Effects of conflict were present in all tasks where the auditory stimulus was task-relevant (content discrimination task II, location discrimination task, and volume oddball). In both auditory discrimination tasks, we observed longer reaction times (RTs) (left bar) and increased error rates (right bar) for incongruent compared to congruent trials. For the volume oddball, we did not observe an effect in RT, but increased sensitivity (d’) on incongruent compared to congruent trials. Dots represent individual participants. The data that is shown here can be found in Figure 4—source data 1. (B) Multivariate classifier accuracies for different stimulus features (auditory congruency, auditory content, and auditory location). Classifier accuracies (area under the curve [AUC]) are plotted across a time-frequency window of −100 ms to 1000 ms and 2–30 Hz. Classifier accuracies are thresholded (cluster-based corrected, one-sided: ${\displaystyle \overline{X}}$>0.5, p<0.05), and significant clusters are outlined with a solid black line. The dotted box shows the predefined ROI on which we performed a hypothesis-driven analysis. Note that the data shown for the volume oddball task was merged over both runs. *p<0.05, **p<0.01, ***p<0.001; n.s.: p>0.05.

Figure 4—figure supplement 1. Effects of exposure to conflict inducing task on behavioral effects of conflict and decoding performance in the volume oddball task of experiment 2.

(A–D) We performed 2 × 2 repeated measures ANOVAs on (A) reaction times (RTs), (B) perceptual sensitivity (d’), (C) hit rates, and (D) false alarm rates with the factors run number and congruency of the auditory stimulus. In (A–D), data are plotted as conflict effects (incongruent – congruent) and for separate runs. The top horizontal line shows significance of the interaction between session and congruency, and markers above the bars indicate significance of paired sample t-tests comparing incongruent and congruent for each run (shown data and results of t-tests can be found in Figure 4—figure supplement 1—source data 1). (A) RTs were unaffected by auditory congruency and run number (statistics in Results). (B) There was no interaction effect between congruency and run number on perceptual sensitivity (d’; statistics in Results), and post-hoc paired sample t-tests (incongruent – congruent) revealed that the effect of congruency on d’ was present during both runs. (C) The interaction between congruency and run number was not significant (F(1,23) = 2.99, p=0.10, ${\eta }_p^2$ = 0.12, BF₀₁ = 1.64), showing that the effect of conflict on hit rate was not different for both runs, although the effect of conflict was present during the first, but not second run. (D) False alarm rates were not modulated by the interaction between congruency and run number (F(1,23) = 0.00, p=0.99, ${\eta }_p^2$ = 0.00, BF₀₁ = 3.45), showing that the effects of conflict were not different between runs. This conflict effect was present during both runs. (E) There were no clusters for which the difference in decoding of all features between the two runs of the volume oddball task was significant, and there were also no differences within the preselected ROI (congruency: t(22) = 0.07, p=0.95, d = 0.01, BF₀₁ = 4.56; content: t(22) = 0.64, p=0.53, d = 0.13, BF₀₁ = 3.81; location: t(22) = –1.25, p=0.22, d = –0.26, BF₀₁ = 2.29), suggesting that processing of these features was not affected by training. Thresholded (cluster-based corrected, p<0.05) accuracies are depicted across the frequency range (2–30 Hz). Plots show the difference in classifier accuracy between the two runs (run 2 – run 1) of stimulus congruency (left panel), stimulus content (middle panel), and stimulus location (right panel) in the volume oddball task. n.s.: p>0.05.

Figure 4—figure supplement 2. Sensory feature decoding in the time-domain.

We trained classifiers on either sound content (A) or sound location (B) in order to see how neural representations of sensory processing were affected by our manipulation of task relevance of these features. (A) Sound content could be decoded from most tasks (except the horizontal random dot-motion [RDM]), and decoding accuracies for sound content were highest for the task in which sound content was the task-relevant feature, that is, content discrimination task II. Decoding accuracies were higher for content discrimination task II as compared to the other three tasks (difference start location discrimination: 328 ms; horizontal RDM: 313 ms; volume oddball: 344 ms). (B) Sound location could be decoded from all tasks. Again, the task in which the decoded feature was task-relevant, that is, the location discrimination task, showed the highest decoding accuracies. Location decoding performance was improved for the task in which this feature was task-relevant (i.e., location discrimination task) as compared to the other tasks. These differences started from 250 ms (vs. content discrimination task II), 234 ms (vs. horizontal RDM task), and 266 ms (vs. volume oddball task). Shaded areas represent the SEM. Bold traces indicate that feature decoding was significantly (cluster-corrected, one-sided t-test, ${\displaystyle \overline{X}}$>0.5, p<0.05) above chance. Horizontal black lines at the bottom depict where feature decoding is significantly different (cluster-corrected, two-sided t-test, p<0.05) between the task in which the feature was task-relevant versus where it was task-irrelevant. CD II: content discrimination task II; HRDM: horizontal RDM task; LD: location discrimination task; VO: volume oddball detection task.