Memory-driven capture occurs for individual features of an object

The visual environment contains abundant information, necessitating selection of relevant inputs for perception and for online storage in working memory (WM). Selective attention enhances detection, discrimination, and processing and can be directed to objects or specific features (feature-based attention), improving behavior and neural responses to task-relevant features while suppressing irrelevant ones. Selection also operates on internal representations in WM, where retro-cues can prioritize or deprioritize objects or their features, yielding benefits for cued feature dimensions and costs for uncued ones. Attentional selection and WM interact: memory-driven capture occurs when distractors matching WM contents involuntarily attract attention during search, but capture can be reduced or abolished if a representation is retro-cued as no longer relevant. Despite object-level evidence for such interactions, it is unclear whether they occur at the feature level. This study tests whether prioritizing specific features of an object in WM modulates memory-driven capture for matching features during visual search, predicting reduced or absent capture by features rendered irrelevant by a retro-cue.

Prior work shows that attention can be directed to features (feature-based attention) to enhance processing of relevant attributes and suppress irrelevant ones (e.g., selective enhancement in visual cortex). Within WM, retro-cues can prioritize items or dimensions, improving memory for cued features and harming uncued features, suggesting feature-level representations can be unbound and selectively modulated. Interactions between external attention and internal selection have been observed at the object level: distractors matching WM content capture attention, but capture is mitigated when WM content is deprioritized via retro-cues. However, whether such bidirectional interactions occur at the feature level had not been directly tested. Differences in feature salience and distinctiveness (e.g., color versus orientation) also influence attentional guidance and might shape memory-driven capture effects.

Two experiments used a dual-task paradigm combining a WM task (remember color and orientation of a single isosceles triangle) with a subsequent visual search. Experiment 1: - Participants: N=18 (18 females; mean age 20.9±2.28), NYU Abu Dhabi students with normal or corrected vision; IRB approved. - Apparatus/Stimuli: Psychtoolbox/Matlab, 22-inch 144 Hz monitor at 57 cm, gray background. Memory item: central isosceles triangle (angles 30°, 75°, 75°; sides 1.46°, 2.84°, 2.84° visual angle). Orientation randomly selected from 180 values (2°–360° in 2° steps). Color from 180 equiluminant CIE L*a*b* values (L=54, a=18, b=−8, radius 59). Retro-cue words: “color”, “orientation”, or “both” (100% valid). Search display: four triangles on an imaginary circle (radius 3.50°), each differing by at least 36° in color and orientation; each had a square gap; target had a large gap (0.66°); distractors had smaller gaps (0.33°). Continuous report used a circular response wheel (180 values) for color or orientation. - Procedure: Trial sequence: memory item 1000 ms; blank 500 ms; retro-cue 500 ms; blank 1000 ms; search array until response. Search conditions: color-match (one distractor shared memory color), orientation-match (one shared memory orientation), non-match (no features matched). Task: speeded mouse click to localize the target (largest gap), response counted if within target quadrant. After 200 ms, memory report screen: adjust and report the cued feature (on “both” trials, color or orientation randomly chosen). 44 trials per condition; total 396 trials, in 12 blocks of 33; 9 practice trials. - Data treatment: Excluded incorrect search trials from RT analyses; removed RTs <150 ms or >3000 ms, and trials >3 SD above mean (2.38% removed). Memory performance quantified as absolute error (degrees of feature space). Experiment 2: - Participants: N=18 (14 females; mean age 22.8±2.31), NYU Abu Dhabi; IRB approved. - Apparatus/Stimuli: As in Exp. 1, except retro-cues were single letters “C” or “O” (smaller font) presented at one of the four future search-item locations to enforce focal attention; no “both” condition. - Cue-location manipulation: On 40% of trials, the cue appeared at the eventual target location; on 60% at a distractor location. On match trials, the cue at a distractor location was always at the matching distractor; on non-match trials, at a random non-matching item location. - Procedure: Same as Exp. 1 aside from cue location. Memory performance for remaining conditions reported in Supplemental Material. - Data treatment: Same exclusion criteria as Exp. 1; search accuracy ~99%. Analyses examined effects of Cue (color/orientation), Match (color-match/orientation-match/non-match), and Cue Location (target/distractor).

Experiment 1: - Memory: Retro-cues improved memory for the cued feature. - Color: Main effect of Cue (color vs both), F(1,17)=6.99, p=0.017, η²=0.29; lower error when color was cued (M=17.33°) than cue-both (M=19.82°). No effects of Match or interaction. - Orientation: Main effect of Cue (orientation vs both), F(1,17)=5.90, p=0.027, η²=0.26; lower error when orientation was cued (M=12.44°) than cue-both (M=14.00°). No effects of Match or interaction. - Search (accuracy 97–98%; 2.38% RT outliers removed): - ANOVA: Cue effect F(2,34)=8.52, p<0.001, η²=0.33; Match effect F(2,34)=17.11, p<0.001, η²=0.50; Cue×Match interaction F(2,34)=2.93, p=0.027, η²=0.15. - Color capture: Present when color relevant. • Cue-color: color-match M=908 ms vs non-match M=833 ms, t(17)=4.32, p<0.001, d=1.02 (capture=75 ms). • Cue-both: color-match M=864 ms vs non-match M=825 ms, t(17)=3.97, p<0.001, d=0.94 (capture=39 ms). • Cue-orientation: no significant color capture, color-match M=842 ms vs non-match M=824 ms, t(17)=1.29, p=0.215, d=0.30. • Capture magnitude difference cue-color vs cue-both was marginal, t(17)=1.90, p=0.075, d=0.45. - Orientation capture: Not observed in any cue condition (all p≥0.358). Experiment 2: - Search (accuracy 99–100%): - With Cue Location factor: Match effect F(2,34)=39.37, p<0.001, η²=0.70; Cue×Match F(2,34)=6.95, p=0.003, η²=0.29; no main or interaction effects of Cue Location (all F<2.98, p>0.101). Data collapsed across locations for main tests. - Collapsed ANOVA: Match effect F(2,34)=39.38, p<0.001, η²=0.70; Cue×Match F(2,34)=6.94, p=0.003, η²=0.29; Cue main effect ns, F(1,17)=2.99, p=0.102. - Color capture: Present under both cues, larger when color cued. • Cue-color: color-match M=914 ms vs non-match M=833 ms, t(17)=7.25, p<0.001, d=1.71 (capture=81 ms). • Cue-orientation: color-match M=856 ms vs non-match M=828 ms, t(17)=4.46, p<0.001, d=1.05 (capture=44 ms). • Capture magnitude larger when color was cued, t(17)=2.93, p=0.009, d=0.69. - Orientation capture: Selective to relevance. • Cue-orientation: orientation-match M=856 ms vs non-match M=828 ms, t(17)=4.46, p<0.001, d=1.05 (capture=28 ms). • Cue-color: orientation-match M=847 ms vs non-match M=833 ms, t(17)=1.15, p=0.266, d=0.27 (ns; capture=13 ms). • Difference in orientation capture between cue conditions not significant, t(17)=1.10, p=0.287, d=0.26. - Overall: Color reliably captured attention and was modulated by feature relevance; orientation captured attention only when made relevant and when attention was pre-focused by a spatially localized retro-cue.

Findings show that feature-specific prioritization within WM modulates memory-driven attentional capture. In Experiment 1, retro-cuing a feature improved its memory precision and eliminated capture by color when color was marked irrelevant, indicating that internal selection alters external attentional guidance at the feature level. Orientation did not capture attention in Experiment 1, likely due to lower salience or difficulty extracting orientation under a broad attentional window. In Experiment 2, a spatially localized letter retro-cue pre-focused attention, yielding robust color capture and revealing orientation-driven capture when orientation was task-relevant. Capture effects did not depend on whether the cue appeared at a target or distractor location, consistent with a comparative decision process requiring inspection of multiple items and with the idea that capture may reflect slower disengagement from memory-matching distractors. Together, results argue for reciprocal interactions between WM prioritization and perceptual attention at the feature level and suggest feature salience and attentional focus shape whether WM features guide attention.

This work demonstrates that internal, feature-specific prioritization within a single-object WM representation modulates memory-driven attentional capture at the feature level. Color consistently captured attention and was enhanced when color was prioritized; capture by color disappeared when color was irrelevant. Orientation captured attention only when it was relevant and when attention was narrowly focused before search. These findings extend object-level interactions between WM and attention to the feature level, supporting models where WM representations can be decomposed into features that differentially interface with attention. Future research should disentangle mechanisms underlying retro-cue benefits (protection from interference, active removal, or mitigation of time-based decay), probe the memory states (active vs latent) of uncued features, examine how feature salience and distinctiveness modulate capture across feature spaces, and increase power to clarify cue-based modulation of weaker orientation capture.

- Orientation-based capture was weak or absent in Experiment 1 and only modest in Experiment 2, limiting inferences about cue modulation; greater statistical power may be needed to detect small effects. - Feature salience/distinctiveness differed (color likely more salient than triangle orientation), potentially biasing capture. - Experiment 2 did not include a “cue-both” condition, preventing direct within-experiment assessment of retro-cue benefits on memory performance. - The spatial letter cue may have induced focused covert/overt attention or eye movements, altering perceptual processing relative to central cues. - The search task required comparative judgments (largest gap), encouraging inspection of multiple items and possibly diluting cue-location effects. - Absence of capture for irrelevant features does not prove removal from WM; uncued features may persist in latent states not directly measured here. - Slight imbalance in cue-location trial proportions (60% distractor, 40% target) could, in principle, influence strategies, though authors argue impact is minimal.