BrainDial: Brain Dynamics of Audiovisual Dialogue

Language comprehension naturally occurs in face-to-face dialogues. During dialogue interactions, language unfolds rapidly, yet comprehension remains possible because listeners anticipate what speakers will say next. Studies using short sentences and isolated visual cues (e.g., lip movements, hand gestures) suggest that these predictions rely on both linguistic content and the speaker’s visual behavior. However, little is known about how the brain processes multiple co-occurring audiovisual cues during natural dialogue, the primary context for language use. Moreover, while dialogues are rich in pragmatic and prosodic audiovisual cues that support comprehension, their influence on word-level predictions remains unclear. The complexity of dialogues has traditionally made it challenging to study their neural basis, but recent methodological advances now allow for the naturalistic investigation of continuous speech processing. For the firts time, BrainDial (Brain Dynamics of Audiovisual Dialogue) will apply these advances in an ecologically valid audiovisual dialogue setting. We will record natural dialogues and analyze their acoustic, linguistic, and visual cues. These annotated recordings will be used in a magnetoencephalography (MEG) study to reveal the fine-grained temporal dynamics and corticoanatomical details of how the brain generates linguisitc audiovisual predictions in real time. By integrating multimodal dialogue corpus annotations, computational linguisitc metrics derived from large-language models (LLMs), body-motion tracking, MEG, and neural computational modelling, BrainDial aims to provide groundbreaking insights into the neural mechanisms of dialogue comprehension.