Why do predicates like 'know' embed both declarative and interrogative clauses, whereas closely related ones like 'believe' only embed the former? The standard approach following Grimshaw (Linguistic Inquiry 10:279–326) to this issue has been to specify lexically for each predicate which type of complement clause it can combine with. This view is challenged by predicates such as 'be certain' which embed interrogative clauses only in certain contexts. To deal with this issue, this paper proposes (i) a novel unified semantics for declarative and interrogative embedding and (ii) a theory where embedding is constrained by semantic considerations. The reason for the apparent unembeddability of an interrogative clause under a given predicate is the resulting trivial meaning of the sentence. Such triviality manifests itself in unacceptability, and crucially it is affected by both the lexical meaning of the predicate and the polarity of the sentence as a whole.