[PS-3.6]ERPs index real-time extraction of statistical regularities in an artificial language speech stream

Noonan, N. 1, 2 , Archibald, L. 1, 2 & Joanisse, M. 1, 2

1 The University of Western Ontario
2 The Brain and Mind Institute

Humans can use distributional statistics to segment fluent speech. Here, we examined this process in real time using event-related potentials (ERPs). Listeners were exposed to 21 minutes of an artificial language (e.g.: Saffran et al., 1996) as cortical EEGs were recorded at 32 electrode sites. An N100 modulation was evident in response to word-onset syllables within the 1st minute of exposure, but attenuated by the 3rd minute perhaps reflecting a novelty response to the auditory stimuli. In response to the more statistically constrained word-final syllables, a fronto-centrally distributed P200 component emerged by the 3rd minute of exposure. This P200 may indicate perceptual segregation or extraction of the statistical regularities between syllables. Examination of ERPs to exposed words post-exposure revealed a centrally distributed P200 component in response to trained words but not foil words, suggesting implicit identification of newly segmented words. Interestingly, the amplitude of the P200 was not related to performance on behavioural responses during this post-test, indicating that explicit measures of statistical language learning are not always a robust indicator of implicit perceptual learning. Taken together, these findings provide evidence of extraction and identification of statistical regularities after minimal exposure to an artificial language.