In this talk, I will discuss three recent "non-reductive" approaches to the semantics of question-embedding, and compare them based on empirical and conceptual arguments. A semantic theory of question-embedding is referred to as "reductive" (after George 2011) if it entails that the interpretation of "x Vs Q" (where x is an individual, V a responsive question-embedding predicate, and Q an interrogative complement) is sufficiently described by the interpretation of a sentence of the form "x Vs p” (where p is an answer to Q, under some notion of answerhood). After introducing George's empirical arguments against reducibility employing factive predicates, I review three recent non-reductive theories of question-embedding, i.e., that of George (2011), Uegaki (2015) and Theiler, Roelofsen and Aloni (2016). These theories are compared in view of two further empirical phenomena, i.e., the behavior of predicates of relevance (Elliot, Klinedinst, Sudo and Uegaki 2016) and the behavior of nominal complements of attitude predicates (Uegaki, to appear). Finally, I will also consider the possibility of comparing the theories in terms of theory restrictiveness, as discussed by George (2011).