Abstract
People fail to individuate two similar objects separated by a brief interval (repetition blindness), and miss the second of two different objects separated by 300 msec (attentional blink). Might these phenomena arise from rational expectations about the rate at which new objects appear? A prior over object transitions can be derived from many world distributions; I use intersaccadic dwell times (lognormal with mean of 100 msec and SD of 3 dB). Conditioned on seeing an object at time t, this prior on object transition yields two hazard functions: the probability that the same object remains in view, and the probability that a new object has appeared. These expectations yield repetition blindness when the probability of seeing the same object exceeds that of seeing anew object: <280 msec post first stimulus, and the attentional blink arises when the observer expects to be looking at neither the same object nor a new object. This model also predicts that perception of the second target will show systematic intrusions. When two objects occur in quick succession, report of the second object identity will include intrusions of distractors preceding the object; however, when the second objects occurs after a longer interval, intrusions will be biased in favor of distractors following the second object. Furthermore, the transition between early and late intrusions on the second target will coincide with a broader distribution of intrusions of all items around the second object. Both of these effects along with their time-courses were born out in previous experimental work. Altogether, a parsimonious ideal observer model based on event-transition priors, which may be derived from intersaccadic dwell times, predicts repetition blindness and attentional blink effects, as well as delay and diffusion during the attentional blink.
ONR MURI N00014-07-1-0937.