Maintaining Common Ground in Dynamic Environments

Abstract Common grounding is the process of creating and maintaining mutual understandings, which a critical aspect sophisticated human communication. While various task settings have been proposed in existing literature, they mostly focus on common ground under static context ignore them overtime dynamic context. In this work, we propose novel setting to study ability both environments. Based our minimal formulation, collected large-scale dataset 5,617 dialogues enable fine-grained evaluation analysis dialogue systems. Through analyses, highlight challenges introduced setting, such as usage complex spatio-temporal expressions create maintain ground. Finally, conduct extensive experiments assess capabilities baseline system discuss future prospects research.