Semantic Representations of Speech Predict Distributed Cortical Activity Related to Language Processing

Abstract

Human’s have a unique ability to produce and comprehend complex language. Language provides a mechanism whereby abstract and concrete thoughts and ideas may be shared or transferred between two individuals in order to bring about common understanding. The structure of natural language allows for the encoding of highly complex and specific concepts into an increasingly complex hierarchical combination of a finite number of sounds, words and sentences. The information encoded by the features at each level of this structure is affected (or determined) by the context of that feature relative to its surrounding features. Specific combinations of sounds (internation, phonemes) collectively form words, which are combined to form sentences. This structure functions through statistical relationship between context and meaning. In this paper, I will present my investigation into how this statistical relationship is processed and represented by the brain during naturalistic language comprehension. I utilize a dataset comprised of real time natural speech collected in unison with electrocorticography recordings from epilepsy patients awaiting treatment in NYU hospitals. I will outline my investigation into modeling the spatiotemporal relationship between contextual representations of words and high gamma component of electrocorticography activity, in order predict the cortical representation of words. Distinct differences in spatial and temporal representation were discovered between speaking and listening conditions, revealing the nature of semantic representation in of different modes of speech in the brain. Connectivity analysis also reveal strikingly similar relationships between spatially distinct brain regions lending support to distinct dorsal and ventral processing pathways in language processing. This study also reveals that significant findings about language representation in the brain can be achieved outside of a controlled laboratory setting in a more naturalistic paradigm.