Can language models use subtext in their communication? Can they use common ground to incorporate subtext more effectively? In our new preprint, we study these questions across various domains — from visual communication to story writing games.