From Speech Semantics to Brain Activity-Timescales Are Key in Their Information Transfer

Area of Science:

• Neuroscience
• Computational Linguistics
• Cognitive Science

Background:

• Fluctuating timescales are inherent in natural phenomena like music, movies, brain activity, and speech.
• Human speech exhibits semantic timescales ranging from words to sentences, varying dynamically during conversation.
• The brain's intrinsic neuronal timescales (INT), indicated by temporally correlated activity, are crucial for information processing over time.

Purpose of the Study:

• To investigate the relationship between semantic timescales in speech and the brain's intrinsic neuronal timescales.
• To determine if and how information is transferred from speech's semantic timescales to the brain's neural activity timescales.

Main Methods:

• Utilized a 7 Tesla functional magnetic resonance imaging (fMRI) dataset from the Human Connectome Project (movie-watching task).
• Extracted semantic time-series (sentence similarity via Sentence-BERT, word depth via WordNet).
• Quantified timescales using dynamic autocorrelation window (ACW) analysis and assessed information transfer via Transfer Entropy (TE).

Main Results:

• Semantic time-series (sentence similarity, word depth) showed significant temporal fluctuations.
• Dynamic ACW analysis successfully captured varying timescales in both semantic inputs and the brain's INT.
• Demonstrated significant information transfer (TE) from semantic timescales to the brain's INT, specifically at the timescale level.

Conclusions:

• Speech's semantic timescales play a key role in modulating the brain's intrinsic neuronal timescales.
• Information transfer from semantic input to neural activity is critically dependent on timescale dynamics.
• This study highlights the importance of timescale analysis in understanding brain-language interactions.