Abstract
Detecting a picture in a rapid serial visual presentation (RSVP) on the basis of a verbal title given just before the sequence (e.g., people in a restaurant) has been shown to be surprisingly easy when pictures are presented for about 100 ms (Potter, 1975). In the present study we presented color photographs of a wide variety of scenes at still higher rates, including 50 ms/picture in six-picture RSVP sequences; the sixth picture functioned as a mask and was never the target. Performance given a verbal title was strikingly good: At 50 ms/picture, the hit rate was .71 and the false yes rate was .14, an overall accuracy of .79. Whether or not the subject had detected the target picture, each target-present trial was followed by a forced choice test between two pictures, both of which matched the target title (but only one had been in the sequence). Even with this more severe test, performance was high and significantly above chance at the 50 ms presentation rate: .775 correct. In work in progress, we find that detection performance remains above chance (.725, N = 5) when the rate is 33 ms/picture (so that the whole sequence takes only 200 ms), although now the difficult forced-choice task is close to chance, at .57. This work suggests that it is not only global image statistics that can be picked up at high rates of presentation, but also more specific object and gist information.