Attitudes (e.g., knowledge, memory, emotion, and so on) held of questions must be complete and true. For example, the sentence “Jenny knows who left” has two conditions: a Completeness condition, that Jenny knows an answer that completely addresses the question “who left”, and a false answer (FA-)sensitivity condition, that Jenny has no false belief relevant to “who left”. Completeness is standardly equivocated to exhaustiveness. For example, a complete answer of “who left” should exhaustively specify all the individuals who left. Adopting this assumption, recent works on question embedding treat the FA-sensitivity condition as a scalar implicature of Completeness and derive it via exhaustification (Klinedinst & Rothschild 2011, Uegaki 2015, Cremers 2016, Theiler et al 2016).

However, investigating into mention-some readings of questions with existential modals (e.g., “Who can chair the committee?”), I argue for a non-exhaustive definition of Completeness that unifies mention-some and mention-all readings of questions (Fox 2013). Further, drawing on observations validated experimentally, I argue against the exhaustification-based account of FA-sensitivity: the FA-sensitivity condition is much stronger than what it can be defined as in any exhaustification-based account. It is concerned with all the relevant false answers, including those that can never be complete. This generalization also suggests a non-trivial prediction against the reducibility view: to recover all the relevant false answers, questions under attitudes must be able to supply partitions; hence, attitudes held of a question cannot be reduced to attitudes held of one answer of this question.