An Exploration of Tonal Expectation Using Single-Trial EEG Classification

We use a machine-learning approach to extend existing averaging-based ERP research on brain representations of tonal expectation, particularly for cadential events. We introduce pertinent vocabulary and methodology, and then demonstrate the use of machine learning in a classification task on single trials of EEG in a tonal expectation paradigm. EEG was recorded while participants listened to two-measure chord progressions that established expectation for resolution to the tonic. Cadential events included the tonic; repeated dominant; II; and silence. Progressions were presented in three keys. Classifications were performed on single trials of EEG responses to the cadential events, with the goal of correctly identifying the label of the stimulus that produced the EEG response. Classification of the EEG responses by harmonic function of the cadential endings across keys produced classifier accuracies significantly above chance level. Our results suggest that the harmonic function of the stimulus can be correctly labeled in single trials of the EEG response. We show that single-trial EEG classification can additionally be used to identify task-relevant temporal and spatial components of the brain response. Using only the top performing time ranges or electrodes of the brain response produced classification rates approaching and even exceeding the accuracy obtained from using all time points and electrodes