Optimizing the mutual intelligibility of linguistic agents in a shared world