Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Spectral Encoding Profile for multilevel linguistic predictions

Clément Sauvage¹, Pierre Guilleminot¹, Benjamin Morillon; ¹Université d'Aix-Marseille

Presenter: Clément Sauvage

Speech processing is believed to rely on two types of information. First, the predictions, which are endogenous and flowing down the cortical hierarchy in a top-down manner, and second, the prediction errors, computed as the difference between the effective and the predicted inputs at each stage of the said hierarchy, in a bottom-up flow of information. The putative role of neural oscillations to mediate those signals is still a question of debate. Here we recorded intracranial EEG activity of 45 epileptic patients, while they listened to ecological speech. We used a Large Language Model to extract proxies of both prediction uncertainty and prediction errors at three linguistic levels, phonemes, syllables and words. We found that encoding of prediction errors and prediction uncertainty peaks respectively in high-gamma and beta bands, only in Primary Auditory Cortex and Superior Temporal Gyrus.

Topic Area: Language & Communication

Extended Abstract: Full Text PDF