Improving Military Common Ground Interactions with Computational Model Alignment

Human teams quickly communicate by referencing common ground, beliefs that team members share and know they share, to shortcut explicit statements and provide timely information. Partners can monitor for evidence of understanding during communication and take the other’s perspective, interactively aligning mental models. However, deployed human-machine teaming (HMT) agents do not use common ground flexibly to accommodate evolving team missions. Agents do not have a rich model of their partner’s beliefs, do not fully use evidence of understanding to clarify what is in the common ground, and do not fluently align or change their computational model (knowledge representation, processing, and language usage). We will research computational model alignment with human dialogue partners using research perspectives like theory of mind and interactive alignment. We will document human maintenance of common ground, survey technology mechanisms (models and algorithms) that support human-like common ground and demonstrate a Phase I proof-of-concept capability. The research will advance the science of interactive alignment between agents and humans and will design a detailed, feasible path forward to a Phase II prototype agent based on computational cognitive architecture. The research will let HMT agents communicate more efficiently through common ground in a collaborative setting.