Learning What Is Irrelevant or Relevant: Expectations Facilitate Distractor Inhibition and Target Facilitation through Distinct Neural Mechanisms

Abstract

It is well known that attention can facilitate performance by top-down biasing processing of task-relevant information in advance. Recent findings from behavioral studies suggest that distractor inhibition is not under similar direct control but strongly dependent on expectations derived from previous experience. Yet, how expectations about distracting information influence distractor inhibition at the neural level remains unclear. The current study addressed this outstanding question in three experiments in which search displays with repeating distractor or target locations across trials allowed human observers (male and female) to learn which location to selectively suppress or boost. Behavioral findings demonstrated that both distractor and target location learning resulted in more efficient search, as indexed by faster response times. Crucially, distractor learning benefits were observed without target location foreknowledge, unaffected by the number of possible target locations, and could not be explained by priming alone. To determine how distractor location expectations facilitated performance, we applied a spatial encoding model to EEG data to reconstruct activity in neural populations tuned to distractor or target locations. Target location learning increased neural tuning to target locations in advance, indicative of preparatory biasing. This sensitivity increased after target presentation. By contrast, distractor expectations did not change preparatory spatial tuning. Instead, distractor expectations reduced distractor-specific processing, as reflected in the disappearance of the Pd event-related potential component, a neural marker of distractor inhibition, and decreased decoding accuracy. These findings suggest that the brain may no longer process expected distractors as distractors, once it has learned they can safely be ignored.SIGNIFICANCE STATEMENT We constantly try hard to ignore conspicuous events that distract us from our current goals. Surprisingly, and in contrast to dominant attention theories, ignoring distracting, but irrelevant, events does not seem to be as flexible as is focusing our attention on those same aspects. Instead, distractor suppression appears to strongly rely on learned, context-dependent expectations. Here, we investigated how learning about upcoming distractors changes distractor processing and directly contrasted the underlying neural dynamics to target learning. We show that, while target learning enhanced anticipatory sensory tuning, distractor learning only modulated reactive suppressive processing. These results suggest that expected distractors may no longer be considered distractors by the brain once it has learned that they can safely be ignored.

Keywords: EEG; attention; brain; expectation; inhibition; statistical learning.

Figure 1.

Task design and behavioral findings of Experiments 1–3. A, A trial sequence of Experiments 1–3. In each condition, in each trial of a sequence of trials, participants had to indicate the orientation (left or right) of a target Gabor. In all conditions except one, a distractor (a Gabor that was horizontally or vertically oriented) was concurrently presented. Across a sequence of trials, the distractor location could repeat, the target location could repeat, or target and distractor locations varied across trials. The search display was presented for 200 ms, and participants had 1000 ms to respond. In all conditions, except the target-only variable (Tv) condition, the target was accompanied by a distractor (a horizontally or vertically oriented Gabor). In the target-repeat (DvTr) and distractor-repeat (DrTv) conditions, the location of the target (Tr) or the distractor (Dr) was repeated over trials in a sequence. In the baseline (DvTv) condition, the target and distractor location varied from trial to trial. In the target-only variable condition (Tv), the location of the target also varied from trial to trial. The number of trials in a sequence ranged between 4 and 12, and the number of search locations between 4 and 8 across experiments. Further, the colors of each condition correspond to condition specific colors in subsequent plots. B–D, RTs as a function of condition and trial position for (B) Experiment 1, (C) Experiment 2, and (D) Experiment 3. E, Boxplot showing benefits of distractor location repetition in distractor-repeat sequences in Experiment 3 as a function of trial position contrasted to random distractor location repetitions in baseline sequences. Solid lines inside boxes indicate the mean. Dashed lines indicate the median.

Figure 2.

Topographic power in a range of low frequencies tracks both the location of the target and the distractor. A, Total power CTF slopes tuned to the target location across a range of frequencies and collapsed across all conditions of interest (i.e., DvTv1, DvTv4, DvTr1, DvTr4). All nonsignificant values were set to zero in a two-step procedure. First, each individual data point was tested against zero with a paired-sampled t test. After setting nonsignificant values to zero, data were evaluated using cluster-based permutation. B, Total power CTF slopes tuned to the distractor location across a range of frequencies and collapsed across all conditions of interest (i.e., DvTv1, DvTv4, DrTv1, DrTv4).

Figure 3.

Target repetition increased anticipatory and poststimulus spatial tuning to target locations. All plots represent the CTF slope, which here quantifies the location specificity of the topographic distribution of activity in the alpha band. A, Evoked power CTF slopes tuned to the target location at the first (left) and final (right) trial in baseline (DvTv) and target-repeat sequences (DvTr). B, Total power CTF slopes tuned to the target location at the first (left) and final (right) trial in baseline and target-repeat sequences. Target repetition increased spatial tuning to the predictable target location already in advance of target presentation. A control (dotted black line) analysis showed that this effect cannot be attributed to lingering effects from the preceding trial. Shaded error bars represent bootstrapped SEM (same applies to subsequent figures). Colored bars on the x axis (blue; green) represent time points where conditions differ significantly from 0 after cluster correction (p < 0.05). Double-colored thick lines indicate time points with a significant difference between the respective conditions after cluster correction (p < 0.05).

Figure 4.

Estimated CTFs on the first and last trial of the sequence across three time windows of interest. A, Fitted CTFs tuned to the target location in baseline (blue; DvTv) and target-repeat sequences (green; DvTr). B, Fitted CTFs tuned to the distractor location in baseline (blue; DvTv) and distractor-repeat sequences (red; DrTv). Estimates were based on a fit to an exponential cosine function (for details, see Results). CTFs are shown separately for the first (top row) and final trial (bottom row) in the repetition sequence. From left to right, CTFs are averaged across three windows of interest (i.e., anticipation: −550 to 0 ms; search display: 0–200 ms; response: 200–550 ms). **p < 0.01, significant difference between CTF amplitudes in baseline and repeat sequences. Although there was a significant difference at the final trial position for target location tuned CTF amplitudes both in anticipation and during search, no such differences were observed in distractor location tuned CTF amplitudes.

Figure 5.

Distractor repetition did not change spatial tuning to distractor locations. A, Evoked alpha power CTF slopes tuned to the distractor location at the first (left) and final (right) trial in baseline (DvTv) and distractor-repeat sequences (DrTv). B, Total alpha power CTF slopes tuned to the distractor location at the first (left) and final (right) trial in baseline and distractor-repeat sequences. Colored bars on the x axis (blue; red) represent time points where conditions differ significantly from 0 after cluster correction (p < 0.05).

Figure 6.

Target repetition reduced the amplitude of the target-evoked N2pc. ERPs evoked by targets were computed only using trials where the target was presented on the bottom left or right from fixation, with a distractor on the midline. A, Difference waveforms (contralateral − ipsilateral) revealing the N2pc are shown separately for the baseline (DvTv) and target-repeat (DvTr) condition on the first (1) and final (4) repetition in the sequence. Double-colored thick lines indicate time points with a significant difference between the respective conditions after cluster correction (p < 0.05). Gray thick lines indicate time points with a significant condition difference after baseline correction (p < 0.05). B, Boxplots represent the difference between conditions (DvTv − DvTr) on the first (dashed) and final (solid) trial of a target repetition block within the N2pc window (170–230 ms). Solid lines inside boxes indicate the mean. Dashed lines indicate the median.

Figure 7.

Distractor repetition reduced the distractor-evoked Pd. ERPs evoked by distractors were computed based on trials where the distractor was presented on the bottom left or right from fixation, with a target on the midline. A, Difference waveforms (contralateral − ipsilateral) revealing that the N2pc and Pd are shown separately for the baseline (DvTv) and distractor-repeat (DrTv) conditions on the first (1) and final (4) repetition in the sequence. Double-colored thick lines indicate time points with a significant difference between the respective conditions after cluster correction (p < 0.05). B, Boxplots represent the difference between conditions (DvTv − DrTv) on the first (dashed) and final (solid) repetition within the Pd window (280–360 ms). Solid lines inside boxes indicate the mean. Dashed lines indicate the median.

Figure 8.

Target repetition was associated with a shortening of the representation of the target location within the N2pc time window, as reflected in decoding accuracy. Shown are target-location decoding accuracies of broadband EEG using all 64 electrodes separately for baseline (DvTv) and target-repeat (DvTr) sequences and the first (1) and last (4) trial in a sequence. Colored bars on the x axis (blue; green) represent time points where conditions differ significantly from 0 after cluster correction (p < 0.05). Double-colored thick lines indicate time points with a significant difference between the respective conditions after cluster correction (p < 0.05). Gray thick lines indicate time points with a significant condition difference after baseline correction (p < 0.05).

Figure 9.

Distractor repetition was not associated with a change in the distractor representation, as reflected in decoding accuracy. Shown are distractor-location decoding accuracies of broadband EEG using all 64 electrodes separately for baseline (DvTv) and distractor-repeat (DrTv) sequences and the first and last trial in a sequence. Colored bars on the x axis (blue; red) represent time points where conditions differ significantly from 0 after cluster correction (p < 0.05).