Speakers and listeners are thought to routinely make sophisticated inferences, in real time, about their conversation partner’s knowledge state and communicative intentions. However, these inferences have only been studied in industrialized cultures. Communicative expectations may be language-dependent, as are many phonological, syntactic, and semantic aspects of language. We study pragmatic inference in communication in the Tsimane’, an indigenous people of the Bolivian Amazon, who have little contact with industrialization or formal education. Using a referential communication task and eye-tracking, we probe how Tsimane' speakers use and understand referential expressions (e.g., ``Hand me the cup.'') across contexts. We manipulated aspects of the visual display to elicit contrastive inferences, including whether the referent was unique or part of a set as well as whether members of the same set differed in size or color. Strikingly, in all cases, patterns of behavior and eye-gaze of Tsimane' and English speakers were qualitatively identical, suggesting that real-time inference may be a core feature of human communication that is shared across cultures rather than a product of life in an industrialized society.