Communication critically requires both a) the basic ability to generate and interpret linguistic signals, and b) the ability to infer the beliefs, desires, and goals of others (so called Theory of Mind, ToM). These abilities have previously been shown to be dissociated. For example, individuals with severe aphasia retain the ability to think about others’ mental states. On the other hand, individuals with autism are impaired in their social abilities, but their basic language processing appears to be preserved. Further, fMRI studies have shown that language and Theory of Mind are implemented in separate sets of brain regions. However, much of what people talk about either directly concerns mental states, or requires mental state inference from information about states of affairs, events, and actions. Consequently, communicative success plausibly requires some coordination between these two systems. Across three datasets, I will show that while separable, the language and ToM networks are functionally associated, showing above-chance synchronization of activity during naturalistic cognition, including both rest and story comprehension. I will then discuss an ongoing effort at understanding the contributions of each system by examining their responses to a broad range of linguistic phenomena that go beyond literal interpretation, such as humor, sarcasm, inference, etc. Although some of these phenomena engage ToM brain regions, I will report a study that suggests that at least some kinds of non-literal inference are computed within the language system. Finally, I will present preliminary findings relevant to understanding the nature of the communicative deficit in autism. In particular, prior work has shown that the basic functional architecture of the language and ToM systems appears to be preserved. However, I will show that during naturalistic cognition (resting state) ToM brain regions show less strong synchronization in individuals with autism, while the regions of the language system are synchronized as strongly as they are in neurotypical individuals. If shown to be robust, this reduced synchronization of the ToM system’s activity holds promise to become a biomarker of autism.