Speech tracking may rely on temporal tuning mechanisms that reset low-frequency amplitude modulations of the neural signal to salient changes in the acoustic input. It has been proposed that structure-building operations are also segregated to narrow-band rhythms in the input. To test such a hypothesis, we collected EEG data from 23 Italian native speakers while they listened to two chapters of an audio book in Italian. To control for self-predictability effects of narrowband signals, we adopted a novel information-theoretic approach between sound and EEG envelopes, based on a whitening preprocessing step before computing directional time-lagged dependencies (DIce, Directed Information based on conditional entropy, Daube et al. 2022), and using narrow-band frequency boundaries defined by Highest Density Regions. We find that an HDR-informed approach increases the amount of mutual information extracted from narrowband neural time series. Importantly, we provide strong evidence for polymodality in both speech and structural unit duration distributions, suggesting that they are not segregated into narrowband rhythms, but rather organised into a flexible, polyrhythmic interaction.

Christoph Daube, Joachim Gross, Robin Ince. 2022. A whitening approach for Transfer Entropy permits the application to narrow-band signals. https://arxiv.org/pdf/2201.02461.pdf