Individual differences in distributional learning for speech: What's ideal for ideal observers?

Research demonstrates that efficient speech perception is supported by listeners’ ability to dynamically modify the mapping to speech sounds to reflect cumulative experience with talkers’ input distributions. Here we test the hypothesis that higher-level receptive language ability is linked to adaptation to low-level distributional cues in speech input. Listeners completed two blocks of phonetic categorization for stimuli that differed in voice-onset-time (VOT), a probabilistic cue to the voicing contrast in English stop consonants. In each block, two distributions were presented, one specifying /g/ and one specifying /k/. Across the two blocks, variance of the input distributions was manipulated to be either narrow or wide, reflecting distributions that were relatively more to relatively less consistent, respectively, in terms of how VOT cued the voicing contrast. As predicted by ideal observer computational frameworks, the participants in the aggregate showed steeper identification slopes for consistent compared to inconsistent input distributions. However, the magnitude of learning showed wide individual variability, which was predicted by receptive language ability as measured using standardized assessments. Individuals with poorer receptive language scores showed diminished distributional learning due to a failure to capitalize on consistent input distributions; instead, their perceptual decisions showed instability even the face of acoustic-phonetic certainty.