ML-SPEAK: A theory-guided machine learning method for studying and predicting conversational turn-taking patterns

Predicting team dynamics from personality traits remains a fundamental challenge for the psychological sciences and team-based organizations. Understanding how team composition shapes team processes can significantly advance team-based research along with providing practical guidelines for team staffing and training. Although the input-process-output model has been a useful theoretical framework for studying these connections, the complex nature of team member interactions demands a more dynamic approach. We develop a computational model of conversational turn-taking within self-organized teams that can provide insight into the relationships between team member characteristics and team communication dynamics. We focus on turn-taking patterns between team members, independent of content, which can significantly influence team emergent states and outcomes while being objectively measurable and quantifiable. As our ML-SPEAK model is trained on conversational data from teams of given trait compositions, it can learn the relationships between individual traits and speaking behaviors and predict group-wide patterns of communication based on team trait composition alone. We first evaluate the performance of our model using simulated data and then apply it to real-world data collected from self-organized student teams. In comparison to baselines, our model is more accurate at predicting speaking turn sequences and can reveal new relationships between team members' traits and their communication patterns. Our approach offers a data-driven and dynamic understanding of team processes. By bridging the gap between individual characteristics and team communication patterns, our model has the potential to inform theories of team processes and provide powerful insights into optimizing team staffing and training. (PsycInfo Database Record (c) 2025 APA, all rights reserved).