Statistical regularities of task-irrelevant dimensions impact auditory decisions

Abstract

Listeners build up statistically driven expectations of what they will hear; however, there is no consensus on how these statistics influence perception, attention, and behavior. Here, we manipulate two statistical properties: global probability (the likelihood of single ‘sound events’) and predictiveness (how often does one sound precede another). We ask how the probability and predictiveness of different acoustic frequencies affect performance on two paradigms where frequency is task-irrelevant: suprathreshold duration identification and near-threshold tone-detection-in-noise. We found that duration decisions are faster and detection decisions are more accurate for high-probability tone frequencies, compared to low-probability tone frequencies. Moreover, when a preceding “cue” tone’s frequency predicts that of a subsequent “target” tone, listeners are faster at judging the duration of the target tone. This latter effect is not solely a result of temporal cueing, as target responses are not facilitated if a cue does not predict the target tone’s frequency. Blending these paradigms to examine the same global and transitional probabilities across duration and detection decisions suggests that statistical learning shapes attention to perceptual dimensions, even when the dimensions are irrelevant to optimal task performance.