Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Towards an analytical framework for investigating the neurophysiological encoding of social contexts in speech perception
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Emily Y.J. Ip1,2,3, Aoife Igoe4, Conor Thornberry2,5, Benjamin R. Cowan3,6, Giovanni M. Di Liberto1,2,3; 1School of Computer Science and Statistics, Trinity College Dublin, University of Dublin, Ireland, 2Trinity College Institute of Neuroscience, Trinity College Dublin, Ireland, 3ADAPT Research Ireland Centre, Ireland, 4School of Engineering, Trinity College Dublin, University of Dublin, Ireland, 5School of Psychology, Trinity College Dublin, University of Dublin, Ireland, 6School of Information and Communication Studies, University College Dublin, Ireland
Speech communication is a fundamental aspect of the human experience, enabling us to express thoughts and emotions. Past research has focused on how the human brain processes speech monologues into meaning. However, what remains unclear are the neural mechanisms that underpin the strategies of speech communication in social scenarios. One phenomenon that is particularly important in social scenarios is the ability to tailor what we say to our interlocutor. For example, the same concept may be communicated in different ways when a speaker is addressing a child or an adult, optimizing the success and ease of communication. Similarly, knowledge of the interlocutor and the social scenario at hand (social context) also influences our expectations when listening to speech. However, the neural foundations of encoding this socially-relevant information during speech communication remain unknown. In the present study, non-invasive electroencephalography (EEG) was recorded from participants as they listened to spontaneous, natural speech from podcast dialogues to investigate how the brain processes speech in dyadic speech interactions. We measured the cortical tracking of different speech features, including how previous context and social context each influence next-word expectations, using multivariate temporal response functions (mTRF) analyses. Such measures of lexical predictions were quantified by the large language model (LLM) Mistral. TRF models incorporating local linguistic context and congruent or incongruent interlocutor speaking characteristics as a proxy for social context were then used to isolate neural signatures of socially-informed predictions. Cortical tracking of acoustic and lexical features in dyadic speech was robustly captured using TRFs and Mistral, validating this approach for naturalistic speech despite challenges such as disfluencies. Models of congruent social contexts yielded higher prediction correlations than their incongruent counterparts, with prediction gains maximized at a specific balance of local and social context. This suggests the brain integrates both types of context in a weighted manner. We additionally present preliminary OPM-MEG source reconstruction analyses of the same paradigm, offering insight into the cortical localization of social context encoding during dialogue listening. Altogether we demonstrate a novel framework for isolating the neurophysiological signature of the social context during speech perception, taking the first steps in measuring such phenomena in a simple naturalistic listening experiment. The outcomes of such work could establish a novel avenue for social speech neurophysiology.
Topic Areas: Speech Perception, Computational Approaches