Abstract
The recurrent processing theory proposes that cortical feedback mechanisms serve to match perceptual hypotheses with incoming visual information. Behaviorally, this recurrent processing can be studied using object-substitution masking (OSM: Di Lollo & Enns, 1998; Di Lollo, Enns & Rensink, 2000). Typically, OSM is generated by presenting observers with a brief visual display in which a target is surrounded by four dots (the mask) and observers make a perceptual judgement about the target. The common finding is that accuracy is worse when the mask remains on for 50-100 ms after the target than when the mask either disappears with the target or remains on for longer periods of time. This decline in accuracy (i.e., the OSM effect) is taken as evidence for a recurrent processing mechanism that substitutes the target representation with the trailing mask representation. It is known that OSM can be weakened by manipulations that encourage segmentation of the target from the mask, such as spatial precuing, focal attention, and relational cues. Statistical learning can also produce segmentation effects, spatial cuing, and shifts in attention, suggesting that regularity learning may also be capable of guiding perceptual hypotheses during recurrent processing. Thus, in two experiments, we investigated whether statistical learning could alter object-substitution masking. The first was an online experiment in which we found a typical OSM effect using novel shape stimuli. In the second experiment, we introduced a probabilistic relationship between adjacent shapes. We found that regularity learning weakened the masking effect, in line with the proposal that statistical learning refines perceptual hypotheses via an implicit prediction mechanism.