We seek to create agents that both act and communicate with other agents in\npursuit of a goal. Towards this end, we extend LIGHT (Urbanek et al. 2019) -- a\nlarge-scale crowd-sourced fantasy text-game -- with a dataset of quests. These\ncontain natural language motivations paired with in-game goals and human\ndemonstrations; completing a quest might require dialogue or actions (or both).\nWe introduce a reinforcement learning system that (1) incorporates large-scale\nlanguage modeling-based and commonsense reasoning-based pre-training to imbue\nthe agent with relevant priors; and (2) leverages a factorized action space of\naction commands and dialogue, balancing between the two. We conduct zero-shot\nevaluations using held-out human expert demonstrations, showing that our agents\nare able to act consistently and talk naturally with respect to their\nmotivations.\n