What The Brain Does As We Speak
AbstractLanguage is a defining human behavior and is dependent on networks interactions amongst broadly distributed neuronal substrates. Transient dynamics between language regions that underpin speech production have long been postulated, yet have proven challenging to evaluate empirically. We used direct intracranial recordings during single word production to create a finely resolved spatiotemporal atlas (134 patients, 25810 electrodes, 40278 words) of the entire language-dominant cortex and used this to derive single-trial state-space sequences of network motifs. We derived 5 discrete neural states during the production of each word, distinguished by unique patterns of distributed cortical interaction. This interactive model was significantly better than a model of the same design but lacking interactions between regions in explaining observed activity. Our results eschew strict functional attribution to localized cortical populations, supporting instead the idea that cognitive processes are better explained by distributed metastable network states.