Statistical noise influences online processing of speech sounds: eye movement evidence from Cantonese VOT perception
How does the human brain process highly variable, non-discrete acoustic information in such a way as to comprehend an incoming signal? In particular, how does the degree of variation affect perceptual certainty? In a ‘visual world’ eyetracking study, native Cantonese listeners saw pictures of words beginning with unaspirated /b/ (e.g. 布bou3 ‘cloth’) or aspirated /p/ stops (e.g. 铺 pou3 ‘shop’). All participants heard a bimodal distribution of auditory stimuli. Distribution peaks corresponded to prototypical voice onset times for each of the aspirated and unaspirated picture names. Participants heard either a Wide (high-variation) or a Narrow (low-variation) distribution of auditory stimuli.
Data were analysed using an advanced new technique, Generalised Additive Modelling (GAM). Results showed increased looks to target between 200-400ms after stimulus presentation in the Wide condition, compared to the narrow condition. This was modulated by distance from category boundaries, as well as distance from category means. We are still working on interpretation, but this suggests that increased variation in the input increases processing cost, even for prototypical pronunciations.