This paper takes up some limitations of truth-theoretic semantics connected with the requirement that knowledge of a compositional meaning theory for a language put one in a position to understand any potential utterance in the language. I argue that associating entities, such as properties, relations, and propositions, with natural language expressions is neither necessary nor sufficient to meet this requirement. I develop an account of how a meaning theory may be formulated in terms of a body of knowledge about a recursive truth theory for a language. I consider two objections. The first is that the sort of knowledge said to suffice to enable one to use a truth theory to interpret its object language is insufficient because it fails to offer insight into semantic structure (Hoeltje 2013). I offer a response to this objection. The second is that the approach relies on antecedent competence in expressions known to be systematically related in meaning to expressions in the object language. I concede that this objection is correct and I argue that how ‘that’-clauses function in explicit statements of meaning, which are our ultimate target, shows that antecedent competence in a language plays an ineliminable role in how they give us insight into meaning. I conclude that to break out of the circle of language that traditional approaches leave us in we need to relate words and sentences to the roles they are supposed to play in our communicative activities described in more fundamental terms.