Fig. 5: The influence of distributional learning is consistent with a gain mechanism exhibiting hysteresis.

In Exp 5a-b, mean detection accuracy as a function of acoustic frequency is plotted in blue; in Exp 6a-b, mean response times in the duration-decision task are plotted in orange. The histograms to the left show distributional regularities for each experiment. Marker size scales with tone probability. In each panel, the darker color (dotted line) indicates behavior in the first half of the experiment; the lighter color (solid line) indicates behavior in the second half, when distributional regularities shift. Error bars are standard error of the mean. a Exp 5a establishes detection accuracy across an equiprobable distribution, then shifts to a unimodal distribution centered on 1000 Hz. Detection accuracy improves for the distribution mode with increased probability and decreases for frequencies with decreased probability. b Exp 5a switches from a unimodal distribution centered at 1000 Hz to an equiprobable distribution. Note the hysteresis at 1000 Hz, where detection remains elevated even into the second half of the study. c In Exp 6a, duration-decision times are flat with equiprobable frequencies in the first half. Introduction of a unimodal distribution centered at 1000 Hz leads to faster duration decisions at the mode. d In Exp 6b the unimodal distribution shifts to equiprobable at the study midpoint and response times shift substantially; note that this effect interacts with the frequency-duration bias identified in Exp 2.