Categorization of autocorrelated sequences

Abstract

Numerous studies show that human subjects can extract the structure of a set of sequences of discrete stimuli. The dynamic control tasks used by Berry and Broadbent (1984), on the other hand, contain a continuous aspect. Following Lories, Dubois, Gaussin (1997) the present experiment shows that subjects can perceive and reproduce some statistical aspects of the structure of simple continuous time series. Our subjects learn to discriminate positively and negatively autocorrelated data series and transfer this ability to new series. Typicality and variability effects are found that are compatible with a standard categorization capability.