PhD Project 15
Data driven investigation of intrinsic dynamic brain states underlying language processing
State of the art in sensory, and recently associative, cortex research reveals a detailed topographical representation, with indications that functions are represented in groups of neuronal ensembles.
The goal of this project is to extract structures in language processing from brain activity dynamics, informed by symbolic and high-dimensional representations of recurring and dominant brain events. We will use extensive sets of unlabelled ECoG data from epilepsy patients and use machine learning to identify canonical brain responses occurring during conversations and listening to television. These structures may map upon existing conceptual language frameworks or generate novel avenues for future research.
This project bridges clinical neuroscience with artificial intelligence, translating mathematical principles to the clinical setting, with potential direct applications in the clinic for language brain mapping. It contributes joint development of data mining techniques and new research strategies to obtain greater insight into cortical organization of language.
The first project on encoding of low-level auditory features in the human perisylvian cortex is completed. A linear model was applied to spectrotemporal features of sound to predict ECoG brain responses to continuous, naturalistic speech stimuli, as in a feature film. The results showed that different cortical regions along the perisylvian cortex were tuned to distinct features of the incoming speech signal. Specific low-level sound features mapped on the temporal rates of higher-level language units, such as phonemes, syllables and words. These results were presented at the 2016 OHBM and SfN meetings. The article manuscript is currently under revision.
Currently, a new project is starting, focusing on modeling ECoG brain responses during conversations and task-free watching to television. In this project, a recurrent neural network will be constructed to extract low-level (auditory) and high-level (language) speech representations from the audio signal and use them to model the ECoG responses.