A Novel Approach to Map the Causal Impact of Brain Stimulation on Semantic Processing With Language Models.

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics

Background:

• Noninvasive brain stimulation, like TMS, is used to study semantic cognition.
• Current methods often use broad stimulus categories, limiting insight into individual word processing.

Purpose of the Study:

• To investigate transcranial magnetic stimulation (TMS) effects on semantic cognition at the individual word level.
• To introduce a novel methodology using language models for analyzing TMS data in semantic tasks.

Main Methods:

• Reanalyzed five public TMS datasets using semantic similarity derived from a language model.
• Modeled trial-level response times and correlated them with language model outputs.
• Compared correlations between effective and control stimulation conditions for semantic similarity, word frequency, and word length.

Main Results:

• Language model correlations with response times significantly differed between effective and control TMS conditions across datasets.
• Evidence suggests TMS effects interact with the meaning of individual words.
• Similar patterns were observed for word frequency and length, indicating potential widespread TMS effects.

Conclusions:

• Language models offer novel insights into neurostimulation's impact on semantic processing at the individual word level.
• The proposed methodology can be applied to enhance the analysis of existing and future TMS studies.
• TMS effects on cognition may be more widespread than previously assumed, influencing factors beyond the primary semantic content.