Local and global rhythmic dynamics in small-group conversations

Abstract

Rhythmic information is an essential part of verbal and non-verbal coordination during human conversations. In small group settings, who coordinates with whom can further shape the overall coordination and create unique global rhythmic patterns qualitatively different from the individual rhythms involved. Despite their significance, the local and global characteristics of these rhythmic dynamics have rarely been studied directly. To address this gap, our study investigates the network topology of turn-taking and rhythmic information flow in small group conversations from the AMI corpus. Nodes in these networks represent participants and edges are either the number of sequential speech turns or a significant flow of acoustic rhythm information between them, inferred through Multivariate Granger causality. We calculated density and centralization metrics for the network and centrality metrics for each node. Simultaneously, Recurrence Quantification Analysis was used to characterize the collective acoustic rhythm arising from this web of interactions. Our findings reveal at least two distinct local and global dynamics in different conversations: sparse networks with centralized-turn-taking that lead to more organized collective acoustic rhythms with a slow tempo, and decentralized and dense networks leading to less organized collective acoustic rhythms with a faster tempo. By uncovering these regularities, our results complement previous knowledge on the complexity of interpersonal coordination of non-verbal rhythmic information in small groups.