Listeners store category identity and uncertainty in memory during spoken word recognition, but not acoustic detail

During spoken language processing, listeners maintain gradient information about past speech input in memory, but it is unclear how detailed these representations are. We test this question in an experiment using the speech discrimination paradigm. We focus on two perceptual decision-making effects: (1) more distant stimuli are more accurately discriminated; and (2) people are more confident about accurate choices, and this relationship is stronger as stimulus distance increases. We consider three types of stimulus ‘distance’ as a proxy for level of representational detail: (a) distance in acoustic cue space; (b) distance in category certainty space; and (c) difference in category identity. We assess which of these levels drive decision-making effects, finding that only category identity and category certainty are significant; difference in acoustic space was not a driver of perceptual decisions. This suggests that during real-time processing, listeners can maintain uncertainty about past speech, but cannot access its acoustic details.