Probability cueing of singleton-distractor regions in visual search: the locus of spatial distractor suppression is determined by colour swapping

ABSTRACT

Observers can learn the likely locations of salient distractors in visual search, reducing their potential to cause interference. While there is agreement that this involves suppression of frequent distractor locations, the results are mixed regarding the stage of suppression: the search-guiding priority map or, respectively, the distractor-defining dimension. Critical for deciding this question is whether or not a distractor-position effect (reduced interference by distractors at frequent locations) is accompanied by a target-position effect (slowed response times to targets at frequent locations) when the distractor is defined in a different dimension to the target: priority-map based suppression would impact target as well as distractor signals; distractor-dimension-based suppression only distractor signals. To identify the factors that are critical for observing one or the other effect pattern, the present study adopted a paradigm in which the distractor was likely to appear in a larger region of the display and orthogonally varied display density (singleton saliency) and random swapping of the distractor and non-distractor colours. Both effect patterns were found consistently, with the critical factor being colour swapping/consistency: with swapping colours, observers tend to adopt a priority-map-based suppression strategy; with consistent colours, a dimension-based strategy.

KEYWORDS:

• Search guidance
• attentional capture
• statistical (distractor location) learning
• distractor suppression

Acknowledgement

This work was supported by German Research Foundation (DFG) grants MU773/16-1, awarded to HJM and ZS, and MU773/14-1, awarded to HJM, as well as a China Scholarship Council (CSC) award to BZ.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Fredrik Allenmark http://orcid.org/0000-0002-3127-4851

Notes

1 However, there was a target-location effect (on distractor-absent trials) when the additional singleton was defined in the same dimension as the target (same-dimension distractors). Sauter et al. (2018, 2019) took this dissociation to argue that learned distractor suppression is implemented at different levels in the hierarchical architecture of search guidance with different- vs. same-dimension-distractors – see below.

2 “Down-weighting” means scaling the “bottom-up saliency” of the distractor by some inhibitory weight, yielding the effective “selection saliency” (see Zehetleitner, Koch, Goschy, & Müller, 2013), where the acquired inhibitory weight is greater for the frequent vs. the rare distractor locations (see Sauter et al., 2018). Accordingly, the effect of inhibiting a distractor signal depends on how bottom-up salient a distractor is, and inhibition is a matter of degree rather than ever absolute (see, e.g., Müller et al., 2010).

3 Suppression of the distractor at the likely location might conceivably also operate at the featural level, that is: the coding of distractor features might be inhibited directly, reducing their potential to generate feature contrast in the distractor dimension (e.g., Gaspelin & Luck, 2018a). In this case, too, one would not expect a target-location effect. We come back to the issue of dimension- vs. feature-based distractor suppression in the General Discussion.

4 This was the case at least initially. However, after extended practice on the task, they changed to a dimension-based suppression strategy, characterized by a distractor-location effect unaccompanied by a target-location effect.

5 In addition to abolishing colour swapping, another difference between the present experiment and that of Wang and Theeuwes (2018a) was that the target appeared equally often at each position even on distractor-present trials. A recent study by Failing, Wang, and Theeuwes (2019) showed that, while the target position distribution does not influence the distractor position effect, an unequal target distribution can result in a target position effect: slower RTs when the target appears at a rare location. The target position distribution could therefore have been another factor potentially explaining why Wang and Theeuwes (2018a) found a target position effect. We showed, in a separate experiment (Zhang et al., 2019), that when this distribution was made equal without abolishing colour swapping, there was still a significant target position effect in the first session (of 1500 trials), but not in the second session.

6 Note the target definition was also more complex, and less certain, in the Wang-&-Theeuwes paradigm, with random swapping, across trials, of the target and non-target shapes (the latter including the distractor); this compares with an unpredictable (left vs. right) target tilt, but with reference to constant vertical non-targets. However, we did not have any intuition as to why this factor might be important. That said, given that target swapping is known to increase distractor interference (e.g., Pinto, Olivers, & Theeuwes, 2005; Lamy & Yashar, 2008; Burra & Kerzel, 2013), a reviewer surmised that the high interference might be responsible for the target-location effect. At variance with this, however, Wang and Theeuwes (2018b) observed target-location effects even with minimal distractor costs.

7 Note that, in both display-density conditions, the stimulus sizes had be to reduced by 75% relative to those used by Wang and Theeuwes (2018a) and Zhang et al. (2019) in order to realize the dense displays without overlapping of the shapes on and across the three rings. To pilot these changes and to ascertain that the target is more salient and, thus, detected and responded to more efficiently in dense than in sparse displays, we conducted a within-participant “baseline” experiment (with N = 10 observers) in which the search displays never contained a distractor, but only an odd-one-out shape target (with the shape assignment to target and non-target items varying randomly across trials), under blocked and counterbalanced color-swapping and no-swapping conditions. As expected, the results depicted a significant main effect of display density (755 [dense] vs. 930 ms [sparse]; (F(1,9) = 15.9, p = .003, ${\eta }_p^2$ = .64, BF > 1000), whereas the effect of colour swapping was non-significant (F(1,9) = 0.38, p = .55, BF = 0.33)).

8 See the Appendix for a table with the actual distribution of target positions produced by this algorithm. Importantly, even when considering only the first 960 trials per group, as we did in most of our analyses, the target appeared very nearly equally often in the top and the bottom region, regardless of which region was the frequent distractor region (50.2% of targets in the bottom region when the bottom region was the frequent distractor region and 50.3% when the top region was the frequent distractor region).

9 In the two “color swap” groups, participants actually performed more than 960 trials, namely, 1440 trials in total. However, to ensure comparability across the four experimental conditions, we analyzed only the first 960 trials per group in all analyses in which all four groups were compared. The analysis of color-swapping effects involved only the color-swap conditions, and so for this we used the full set of 1440 trials.

10 Figure 2 suggests that the main effect of display density may have been obscured, to some extent, by a speed-accuracy trade-off, i.e.: nearly error-free responding (only) by the “dense displays, with color swapping” group. Examining for the main effect in an ANOVA on a measure combining speed and accuracy (Balanced Integration Score; Liesefeld, Fu, & Zimmer, 2015; Liesefeld & Janczyk, 2019) yielded: F(1, 116) = 7.48, p = .007, ${\eta }_p^2$ = .061, BF = 5.5.

11 Note that in this analysis (which involves examining for the effects of relatively rare cross-trial transition effects), we included the full available set of 1440 trials (rather than just the first 960 trials) per participant, so as to increase the reliability of the estimates for each condition compared. An equivalent analysis based on the partial data set (of 960 trials) revealed essentially the same pattern, except that the “distractor condition on trial n” effect was not significant.

12 However, the probability cueing effect, and its dependence on colour swapping and display density, was somewhat different between participants who responded correctly to the forced-choice question (“aware” participants) and those who did not (“unaware” participants): awareness interacted significantly with colour swapping (2-way interaction: F(1, 105) = 6.36, p = 0.013, ${\eta }_p^2$ = .064, BF = 2.8) and with both display density and colour swapping (3-way interaction: F(1, 105) = 4.70, p = 0.032, ${\eta }_p^2$ = .043, BF = 1.9)). With dense displays, the cueing effect was larger for aware than for unaware participants, irrespective of colour swapping (colour swapping: 109 vs. 99 ms; no colour swapping: 92 vs. 70 ms). With sparse displays, by contrast, while cueing was pronounced for aware vs. unaware participants in the absence of colour swapping (74 vs. 31 ms), it was less marked with colour swapping (17 vs. 91 ms). There were no significant effects of awareness on overall distractor interference or on the target position effect. Thus, awareness appeared enhance the probability cueing effect generally with dense displays, i.e., under conditions of high (distractor) colour feature contrast. With sparse displays, awareness appeared to enhance cueing under conditions of colour consistency, but to impede it under conditions of colour inconsistency. Apart from the need to replicate this (in the literature atypical) pattern, for interpreting it, it would likely be important to know whether “aware” participants were also aware of the distractor manipulation, which was not tested.

13 Note, though, that Theeuwes (2004) had participants search for a diamond target in displays that contained other unique shape items (one square and one triangle, besides at least two non-singleton, circle, non-targets) to encourage the adoption of a “feature-search”, rather than a singleton-detection, mode (cf. Bacon & Egeth, 1994). Despite this, presentation of an additional colour singleton (distractor) caused significant interference when the display size was rendered large (12 or 20 items) by adding more shape homogeneous non-singleton (circle) items to the search array, with little difference in the magnitude of interference between 12- and 20-item displays. Similar to our results, this could be explained by assuming that the additional filler items increased the saliency of the target and the distractor commensurably.

14 One related issue in this context concerns to what extent what looks like feature selectivity within the colour dimension (e.g., in Failing, Feldmann-Wüstefeld, et al., 2019) is really a form of dimensional selectivity. In the search literature, feature-specific selection/de-selection effects have been demonstrated almost exclusively using color-defined targets/ distractors. However, there is evidence that “color” is special (e.g., D’Zmura, 1991; Found & Müller, 1996; Lindsey et al., 2010; Müller et al., 2003) and may in fact be best conceived as consisting of a number of relatively independent (though coupled) “sub-dimensions” (e.g., Found & Müller, 1996; for review, see Liesefeld, Liesefeld, Pollmann, et al., 2019). In any case, to corroborate general/genuine feature selectivity, at the very least, the critical colour findings would need to be reproduced with other feature dimensions.

Additional information

Funding

This work was supported by China Scholarship Council; Deutsche Forschungsgemeinschaft [grant number MU773/14-1,MU773/16-1].