Proceedings of the Annual Meeting of the Cognitive Science Society

Team science research heavily relies on communication data—that is, data derived from audio, video, or text-chat communication streams between team members. Between transcription and content analysis, significant overhead is required to work with these data. Recent developments in natural language processing (NLP) may help ameliorate time constraints in this domain. Using transcript data, the present study, presented as a proof-of-concept, assesses how the BERT NLP model performs in a team communication categorization task, in comparison to ground truth measures. This work builds upon past work that relied on human-coded transcripts to identify phase transitions in team collaboration. Results suggest BERT’s capabilities at phase change detection are promising for experienced teams, though further iteration is needed on the methods in the current study. Applications of this work extend to real-time collaboration with an artificial agent, as this requires the real-time semantic processing of human communication data.