DyadicTickTacking: Multimodal Co-Evolution of Joint Strategies and Turn-Taking in Expert-Novice Scaffolding

Abstract

Performing a cooperative task requires partners to coordinate their actions, but does the presence of an expert change how this process develops over time? We address this question through DyadicTickTacking, a paradigm where pairs of participants jointly navigate proximity-coupled cursors toward shared targets across 160 trials. Combining unsupervised trajectory clustering with automated speaker diarization and turn-taking analysis, we compare Novice-Expert and Novice-Novice dyads across four learning blocks. Three emergent movement strategies are identified via HDBSCAN, forming a validated skill hierarchy. A permutation test reveals that expert-novice dyads synchronize their strategies trial-by-trial above chance, while novice-novice dyads do not. In parallel, mixed-design ANOVA uncovers divergent turn-taking trajectories: expert-novice dyads progressively reduce speech density and shorten turns, whereas novice-novice dyads show no directional verbal structure. These multimodal findings suggest that expert scaffolding is a temporal process, manifesting as a progressive transfer of control.