HARMONY IN DISCORD: A MULTIMODAL EXPLORATION OF CONSTRUCTIVE BEHA VIOURS FOR MANAGING DISAGREEMENTS

Abstract: Conflict can be dysfunctional, but when managed well, fosters learning, creativity, and understanding. Recent research has shifted toward conflict expressions, yet methodological limitations hinder theory testing and development where dynamic interactions and nuanced behavioral cues are difficult to capture. This study examines conflict expressions—linguistic, paralinguistic, and visual cues that shape emotional responses and interpersonal tension—to understand how conflict can be navigated productively. Using a Super Sabbatical negotiation exercise with 94 participants (47 dyads), where no Zone of Possible Agreement exists, we observe negotiations in a naturally conflictual setting rather than prompting conflict. We applied a Hidden Markov Model (HMM) to segment conversations into four stages—Opening, Testing, Trying, and Deciding—revealing key behavioral differences between high- and low-conflict dyads. High-conflict groups exhibited more back-and-forth in Opening and Trying, while linguistic markers varied dynamically across stages. We leveraged machine learning models to extract linguistic, paralinguistic, and visual cues at the millisecond level (frame-by-frame), capturing subtle patterns in conflict dynamics. Taking advantage of fine-grained data, our abductive approach allows us to explore co-occurring behaviors and let the data guide the development of plausible hypotheses. Our findings suggest that turn-taking dynamics, sentiment shifts, and verbal politeness play key roles in managing conflict productively.

Keywords: Conflict expressions, multimodal data, constructive conflict, machine learning