EEG evidence of the extraction of regularities in the speech signal by human neonates

Fló, A. , Palu, M. & Dehaene-Lambertz, G.

UNICOG, CEA, INSERM U992, NeuroSpin Center, Gif?sur?Yvette, France

Vast literature evidences remarkably infant's abilities for detecting statistical regularities (e.g. Saffran et al. 1996; Teinonen et al 2009; Fló et al. 2019). Nevertheless, how statistical learning operates across different features of the stimulus and what is extracted, of particularly interest in infants, remains unknown. Here we investigated EEG neural correlates of statistical learning in newborns when hearing an artificial language. Neonates heard speech built by random concatenation of syllables, and of four three-syllable pseudo-words. Using a frequency-tagging approach (e.g. Kabdebon et al. 2015) we observed enhance power at the syllabic rate during both the random and structured streams, while at the word rate, only during the structure stream, suggesting that neonates segmented the stream. In an after learning phase, we investigated evoked responses for sequences in isolation adhering in a different extent to the speech structure. The pattern of responses suggested that newborns did not only extracted the TPs, but also encoded the first syllable of the words. In a second set of experiments, we are currently investigating whether neonates detect regularities indiscriminately over any feature of the stimuli by familiarizing with stimuli having structure in two orthogonal features: the phonemic identity or the voice identity of the syllables.