Questions with NPIs Andreea C. Nicolae December 16, 2014 to appear in Natural Language Semantics Abstract This paper investigates how the distribution of negative polarity items can inform our under- standing of the underlying semantic representation of constituent questions. It argues that the distri- bution of NPIs in questions is governed by the same logical properties that govern their distribution in declarative constructions. Building on an observation due to Guerzoni and Sharvit (2007) that strength of exhaustivity in questions correlates with the acceptability of negative polarity items, I propose a revision of the semantics of questions that can explain this link in terms already familiar from the literature of negative polarity, namely the availability of a local downward entailing en- vironment. I argue for a new theory of questions that takes strength of exhaustivity to be encoded internal to the question nucleus rather than in different answer-hood operators (Heim 1994). This switch, while conceptually a simple move, has far-reaching consequences in the domain of ques- tions. I also show how this new analysis can account for a host of issues related to NPIs, such as the subject-object asymmetry noted by Han and Siegel (1997), the interaction between NPIs and high versus low wh-adjuncts, the varying acceptability of NPIs in the restrictor of which-phrases, and the contrast between weak and strong NPIs. 1 Introduction The goal of this paper is to argue for a unified account of the distribution of negative polarity items (henceforth NPIs) in wh-questions, which has posed a problem for both the semantics of questions and theories of NPIs. These items are acceptable in questions, despite the fact that questions do not prima facie share the key properties we find in the other environments in which NPIs surface. Specifically, within current frameworks of question semantics there is no way to argue that questions give rise to downward-entailing inferences, which is what unifies all other NPI environments. Building on an insight by Guerzoni and Sharvit (2007) that the distribution of NPIs in questions is even more constrained than initially thought, namely that NPIs are acceptable only in questions that receive a strongly exhaustive interpretation, I maintain, with others, that current theories of questions are not fine-grained enough to account for the systematic behavior of NPIs in questions, on the one hand, and declaratives, on the other. I will argue that this calls for a re-evaluation of the semantics of questions, particularly with respect to how to encode the ambiguity in strength exhibited in embedded contexts. I will show that via a conceptually minimal switch in the semantics of questions we can not only explain why the distribution of NPIs should correlate with the strength of the question, but we can also do so in a manner that allows us to 1