Ensemble perception involves more than means and standard deviations: Mapping internal probabilities density functions with priming of pop-out

摘要

Observers can estimate summary statistics of visual ensembles, such as the mean or variance of distributions of color, orientation or motion. But attempts to show that the shape of feature distributions is accessible have not been successful. Using a novel "priming of pop-out" paradigm we show for the first time that observers encode not only means and standard deviations of orientation distributions, but also their shape. Observers searched for an oddly oriented line in a set of 36 lines. Within streaks of 5 to 7 trials, distractor orientations were randomly drawn from a pre-defined distribution while target orientation differed by 60 to 120 degrees from the distractor distribution mean. We analyzed RTs on the first trial of each streak by orientation difference between the present target and the mean of preceding distractor distribution (T-PD distance). We replicate effects of distractor heterogeneity and observers' sensitivity to the mean and variance of preceding distributions. Most importantly, however, we demonstrate that repetition effects differ following uniform and normal distributions of identical range. We assume that the higher the subjective probability (learned in a previous streak) that a stimulus with a given orientation is a distractor, the longer the RTs when it is a target. Following a normal distribution, RTs gradually decrease as T-PD increases. Following a uniform distribution, responses are similarly slow when the target falls within or close to the range of a preceding distractor distribution, but only when T-PD further increases do RTs decrease. Distribution shape is, in other words, reflected in the RTs: RTs and hence the corresponding expectations are "uniform" when the preceding distractor distribution is uniform. We conclude that observers are able to encode the shape of stimulus distributions over time and that our novel paradigm allows the mapping of observers' internal probability density functions with surprising precision.