Sequential patterns in turn durations of dynamic speech interaction

People’s conversations follow distinct interaction patterns. We examine the role of time in these patterns by visualizing, analyzing, and identifying patterns in transitions between speakers using tools from dynamical systems theory. We map spontaneous conversational activities as paths in state space, calculate the likelihood of state transitions, and evaluate the correlation between the durations of subsequent turns. Our measurements demonstrate that the duration of consecutive turns follows a structured pattern: consecutive turns are highly correlated in length, except for pairs involving at least one short turn. These results advance descriptive knowledge of turn-taking dynamics and may help inform the design and evaluation of conversational systems.