hide
Free keywords:
-
Abstract:
Much of our understanding of sensory decoding stems from the comparison of human to ideal observer performance in simple two-alternative discrimination tasks. The optimal Bayesian decoding strategy consists of integrating noisy neural responses into a reliable function that captures the likelihood of specific stimuli being present. As only two stimulus values are relevant in a two-alternative discrimination task, the likelihood function has to be read out at two precise locations to obtain a likelihood ratio. Here, we report a new perceptual bias suggesting that human observers make use of a less optimal likelihood read-out strategy when discriminating grating spatial frequencies. Making use of spectrally filtered noise, we induce an asymmetry in the stimulus frequency likelihood function. We find that perceived grating frequency is significantly altered by this manipulation, indicating that the likelihood function was sampled with remarkably low precision. Although observers are provided with prior knowledge of the two relevant grating frequencies on each trial, they evaluate the likelihood of a broad range of irrelevant frequencies. Overall, our results suggest that humans perform estimation of a stimulus variable of unknown quantity rather than evaluation of two known alternatives when discriminating grating spatial frequencies.
