<p>What language is, as a phenomenon, is a common topic of metalinguistic discourse. The current interest in Large Language Models (LLMs) provides an opportunity to examine how this discourse differs across fields of study. I examine the descriptions of language in online educational materials and in a course designed to teach Python and its use in Natural Language Processing (NLP). Instructors and textbooks explicitly introduce language as a semantic, referential, and denotationally transparent system of logical units encoded in text. In practice, instructors use a wide variety of nonreferential indexical communicative strategies such as hedging, voicing, and gesture, as well as semantically ambiguous linguistic techniques like anthropomorphism, metonym, and metaphor, that are not included in the educators’ definition of language. Instead of treating this discourse as evidence of a particular language ideology, I argue it makes sense to speak of domain-specific language ontologies. Using an ontological frame explains why concepts of language across neighbouring disciplines such as computational linguistics, sociolinguistics, and linguistic anthropology often feel incommensurable.</p>