Abstract
Crowding, the failure to correctly identify objects in the periphery when surrounded by nearby distracting forms, continues to be an effective phenomenon for studying general object recognition. Recent studies with simple stimuli have shown convincingly that crowding reveals an averaging process, at least for continuously-varying stimuli. We asked whether crowded letter identification, a multi-dimensional categorical operation, can expose a pattern of responses interpretable as positional averaging. To directly probe this question, we generated symbols composed of a circle and a vertical line, combined to form shapes varying on a spectrum between ‘b’ and ‘d’ or ‘p’ and ‘q’. Our targets consisted of one lower-case Arial letter from the set ‘bdpqol’ in a central position flanked by symbols on either side of the letter. Observers were asked to identify only the central letter, and were informed of the six possible choices. The stimuli were presented at 10 degrees below fixation for 150 ms. When an error occurred (33% of trials), the mistakes were essentially due to the vertical line features. The circle feature rarely migrated from the flankers to the target and was seldom undetectable in a target letter. Instead, the vertical line either migrated from a flanker into the target or shifted its position within the target. Consistent with the positional averaging prediction, the relative position of the vertical line in the flankers biased the reported letter. Observers were twice as likely to respond with an erroneous letter whose ascender/descender corresponded with the average of the relative line positions of the flankers. Our results are also consistent with the “faulty feature migration” explanation of crowding, which states that features from sufficiently similar flankers are inappropriately integrated into the target. Our new contribution is that letter features migrate in a fashion that respects the internal spatial configuration of the flanking forms.
NIH T32-EY007043, R01-EY012810.