Abstract
Purpose: To analyze the role of irrelevant, visually salient boundaries in the short-term storage of visual information. Method: On each trial, the S viewed two briefly flashed, 6-by-6, random checkerboards separated by a 1 sec. ISI and judged (with feedback) whether they were different. In Exp. 1, the S was aware that the difference (if there was one) was always in the contrast polarity of a single square. Performance was observed (in separate blocks) for two task variants: in the homogeneous task, the entire checkerboard was achromatic; in the segmented task, the squares in the lower right quadrant were achromatic, but those in the upper left, upper right, and lower left quadrants were tinted red, green, and blue, respectively. This color scheme remained fixed across all trials in the segmented task, producing irrelevant chromatic boundaries. All dark squares were equiluminant, as were all bright squares. Expt. 2 was analogous to Expt. 1, except that the roles of contrast polarity and color were reversed: i.e., in both tasks S's tried to detect a change in (equiluminant) color; and in the segmented task, the irrelevant boundaries between quadrants were defined by contrast polarity. Results: In both experiments, performance was significantly worse in the segmented than in the homogeneous task. Although the boundaries (either chromatic or contrast-defined) in the segmented task were both irrelevant and fixed from trial to trial, they nonetheless impaired performance. In addition, changes in four central squares were easier to detect than other changes in the homogeneous task but harder to detect in the segmented task (in which these squares abutted two irrelevant boundaries). Conclusions: these findings suggest that (1) at least some of the statistics used to sense changes recruit information from across multiple squares, and (2) these statistics are impaired if they have to pool information across preattentive boundaries.