Abstract
People watching rapid serial visual presentation (RSVP) of scenes may experience binding errors so that objects from different scenes presented in succession are seen as cooccurring within the same scene (with line drawings, Intraub, 1989; pairs of letters, Bowman & Wyble, 2007; and letters in words, Holcombe & Judson, 2007). One might assume that since natural scenes, unlike artificial stimuli, may be processed holistically (e.g., Oliva & Torralba, 2006), they would be less affected by cooccurrence misbinding. To test this assumption we compared two types of stimuli: colored photographs of distinctive people sitting in chairs, and digit-letter pairs.
On each trial two target images were embedded in an RSVP sequence of distractors (photographs of indoor scenes or pairs of keyboard symbols) presented at the rate of 93 ms/image. The second target (T2) followed the first one (T1) at lag 1, 2 or 6. Subjects were asked for an ordered report of both T1 and T2. As expected, there was a substantial attentional blink for both kinds of materials, with lower report of the second pair at lag 2 compared with lag 6. Lag 1 performance was good (lag 1 sparing).
For both natural scenes and alphanumeric pairs, binding and order errors were maximal at lag 1. This loss of temporal information at lag 1 can be considered the cost for successful processing of two scenes presented in close succession (Bowman & Wyble, 2007). Surprisingly, the proportion of binding errors for integrated natural scenes (a person in a chair) was the same as for alphanumeric pairs. For both types of stimuli there were more binding errors than order errors (reversals) for correctly bound pairs. This pattern of results suggests that natural scenes in RSVP are rapidly decomposed into their constituent objects before the processing of temporal order.
Research is supported by MH 047432.