Based on a corpus of conversational English, I argue that the standard view of complements as subordinate clauses in a grammatical relation with a complement-taking predicate is not supported by the data. Rather, what has been described under the heading of complementation can be understood in terms of epistemic/evidential/evaluative formulaic fragments expressing speaker stance toward the content of a clause. This analysis, in which CTPs and their subjects are stored and retrieved as formulaic stance markers accounts for the grammatical, pragmatic, prosodic, and phonological data more satisfactorily than a complementation analysis.