On selection in vision.

Abstract

The discrepancy between the capabilities of perceptual systems and effector systems makes mechanisms necessary that can selectively designate a subset of the perceptual information available to control an action. In vision, since foveation is neither sufficient nor necessary for this selection, central selection mechanisms have to be postulated. This study is concerned with the question of at what level this central selection takes place in one type of task, the partial-report bar-probe task. Two theories are relevant: early selection theories, asserting that central selection operates in a stage containing rapidly decaying precategorical visual information, and late selection theories, arguing that central selection operates in a stage containing abstract categorical information. A prediction following from the assumptions of early selection theories, but not of late selection theories, is that in bar-probe tasks the function relating the number of visual confusions to probe delays has to exhibit an inverted-U shape. This prediction was tested in five experiments, two with colors as items and three with letters. In all experiments, the proportion of visual confusions first increased with increasing probe delays and then decreased with still larger probe delays. The results support the early selection theories for small probe delays. For larger probe delays, late selection has to be postulated.