Decoding speech comprehension from continuous EEG recordings

Abstract

Human language is a remarkable manifestation of our cognitive abilities which is unique to our species. It is key to communication, but also to our faculty of generating complex thoughts. We organise, conceptualise, and share ideas through language. Neuroscience has shed insightful lights on our understanding of how language is processed by the brain although the exact neural organisation, structural or functional, underpinning this processing remains poorly known. This project aims to employ new methodology to understand speech comprehension during naturalistic listening condition. One achievement of this thesis lies in bringing evidence towards putative predictive processing mechanisms for language comprehension and confront those with rule-based grammar processing. Namely, we looked on the one hand at cortical responses to information-theoretic measures that are relevant for predictive coding in the context of language processing and on the other hand to the response to syntactic tree structures. We successfully recorded responses to linguistic features from continuous EEG recordings during naturalistic speech listening. The use of ecologically valid stimuli allowed us to embed neural response in the context in which they naturally occur when hearing speech. This fostered the development of new analysis tools adapted for such experimental designs. Finally, we demonstrate the ability to decode comprehension from the EEG signals of participants with above-chance accuracy. This could be used as a better indicator of the severity and specificity of language disorders, and also to assess if a patient in a vegetative state understands speech without the need for any behavioural response. Hence a primary outcome is our contribution to the neurobiology of language comprehension. Furthermore, our results pave the way to the development of a new range of diagnostic tools to measure speech comprehension of patients with language impairment.