In conversation, much is communicated without being directly said. By leveraging an understanding of how language relates to mental states and processes, communication becomes a window into a speaker’s thinking. In this dissertation, I demonstrate how children (ages 4-to-9) come to readily reason about others’ mental states based not just on what they say, but how they say it—from how easily something is said (Chapter 1), to how surprised someone seems (Chapter 2), and even how someone is spoken to (Chapter 3). In Chapter 1, I explore the humble “um.” While disfluencies in speech are often overlooked as meaningless errors by laypeople and researchers alike, I demonstrate that children interpret disfluencies as socially meaningful—over and above the content of what is said—and use them to flexibly infer a speaker’s knowledge and preferences. In Chapter 2, I explore how children reason about the implications of conversational cues in feedback, specifically how markers of surprisal and production difficulty (e.g., “Oh! Um… Sure”) lead children and adults to infer a speaker’s underlying expectations. I find that conversational feedback not only signals a speaker’s expectations, but also provides an unappreciated avenue for the transmission of social beliefs and stereotypes. In Chapter 3, I show that how someone is spoken to may shape the mental inferences that children make about that person before that person ever says a word. When a speaker offers basic categorical information, children and adults infer that the listener is likely unfamiliar with the topic at hand. Across these three chapters, I argue that children are actively, rationally, and flexibly inferring mental states by integrating subtle conversational cues, context, and prior discourse. Capitalizing on their skills as budding mentalists, children are learning to extract social meaning from subtle conversational cues—skills that are fundamental to becoming smooth conversationalists and sophisticated social learners.