Selective attention is crucial to natural vision, contributing both to the quality of perceptual experience and to the guidance of saccadic eye movements. A central question is how attention is distributed in real tasks, handling the requirements of both perception and saccades concurrently. We studied attention while counting, an active visual task that can be performed either with or without saccades. Subjects counted 10–19 randomly-distributed dots (16' diam) in an 8 deg diam region while a probe stimulus (tilted Gabor patch, σ=.9 deg) was flashed (65 ms) at one of 4 locations during a randomly selected pause between saccades. During counting, identification of the orientation of even high-contrast Gabors was poor, regardless of whether counting was mediated by saccades or by shifts of attention during steady fixation. Errors in orientation identification were not due to failures to notice the Gabor, since detection and localization were accurate. Orientation identification was far better when comparable patterns of saccades were made without the concurrent counting task and without pre-defined saccadic targets. We also tested a pointing task in which the cognitive load was greater than that of simply looking, but less than that of counting. When subjects pointed to 7–15 randomly-distributed dots using a stylus on a graphics tablet, two showed a loss in Gabor orientation identification comparable to that during counting, two showed a modest loss, and one no loss at all. Taken together, these results show a strong attentional bottleneck at the level of perceptual identification. Active tasks, such as counting or pointing (but not necessarily the saccades themselves), place strict limits on what we can identify during any single fixation pause, but modest limits on the ability to detect, locate and move the eye to new or important objects.
NIH EY15522 NSF DGE0549115