Predicting Multi-Party Turn-Taking Using Gaze and Speech Dynamics
August 31, 2026
Logistic regression models trained on the GaMMA corpus classify floor-transfer outcomes as gaps or overlaps in four-person dialogues. The method utilizes gaze features including transition motifs, entropy, and mutual gaze alongside speech cues to interpret interpersonal closeness and conversational transitions.
HOW THIS AFFECTS YOU
●
researcherYou can use these interpretable features to model non-verbal cues in multi-party conversational AI.