Purchase this article with an account.
Maria F. Dal Martello, Laurence T. Maloney, Cynthia Sahm, Lothar Spillmann; How the past gives way to the present: Evidence for Bayesian updating with repeated presentation of ambiguous motion quartets. Journal of Vision 2005;5(8):661. doi: https://doi.org/10.1167/5.8.661.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction: In a motion quartet (MQ)with one repetition, two disks that fall at the endpoints of a diameter of an invisible circle are briefly presented and then replaced by two disks that fall on a second diameter. The observer sees rotation of the first pair into the second. By varying the angle between the diameters, one can estimate a psychometric function P[L|angle]. We reported last year that recent task history exerted a strong effect on perceived direction of motion, shifting the threshold parameter of the psychometric function. In particular, the observer is strongly biased to see motion in the same direction as the motion perceived on the most recent trial. This hysteretic effect (measured as the change in threshold) is little affected by increasing the time interval between trials (to as much as 30 seconds). Purpose: We sought to determine how repetitions of the same motion quartet would affect hysteresis. It is plausible, for example, that multiple presentations of the same sensory information would eventually ‘overpower’ the influence of recent task history, reducing hysteresis to 0. But precisely how? Methods: We measured the hysteresis induced by the previous trial for two subjects who judged 3840 MQ trials each. In different blocks, the MQ was presented 1, 2, 4 or 8 times in each trial. The observer reported the direction of motion of the last presentation only. Results: For both observers we found that hysteresis decreased rapidly with number of repetitions and that a log-log plot of hysteresis versus repetition was approximately linear with slope −1. Doubling the number of repetitions halves the hysteretic effect. This outcome is consistent with a Bayesian updating model where an initial prior probability that the stimulus will go left (based on recent task history) is successively combined with independent likelihood terms, one for each repetition of the stimulus.
This PDF is available to Subscribers Only