Single-trial Based EEG Classification of the Dynamic Representation of Speaker Stance: A Preliminary Study with Representational Similarity Analysis

A classification approach combining machine learning and representational similarity analysis was performed on neurophysiological data (i.e., electrophysiological responses) to distinguish speakers of different stances. The trial-based classification based on a single subject data revealed temporo-spatial neural representations that most accurately predict speaker stance at early and late time windows that roughly corresponded with the ERP (event-related potentials) elicited by each stance in our earlier study (Jiang & Pell, 2015). These findings highlight the significance of applying multivariate pattern recognition to the multimodal data (such as neural responses) to predict dynamic categorical representations relevant to social interaction.