Going back to Ross (1967) and Chomsky (1973), it is recognized that acceptability of wh extraction out of syntactic “island” configurations is ill-formed. One kind of configuration that has received a lot of attention is NP objects. Here, extraction is sensitive to definiteness: Whereas extraction from an indefinite NP is judged as grammatical, extraction from a definite NP is reported to be ungrammatical (e.g., Which book did Sharon read a/*the chapter of?; Fiengo and Higginbotham 1981). However, NPs are not homogeneous (Grimshaw 1990), but minimally comprise (i) concrete (ii) result, and (iii) process nouns. It has been claimed that these three nominal subclasses respond differently to the definiteness asymmetry in NP subextraction (Davies and Dubinsky 2003). We reexamine Davies & Dubinsky’s generalizations using three acceptability judgement experiments, coupled with novel corpus data. We demonstrate the relevance of the distinction between relational nouns (e.g., chapter) and functional nouns (e.g., title). We propose a novel analysis whereby questions with extraction out of a definite object are degraded if uniqueness presupposition of definite the is not satisfied. This extends a growing body of literature arguing that most “island” effects are not rooted in syntax but arise from semantic-pragmatic inconsistencies (Erteschik-Shir 1973; Szabolcsi and Zwarts 1993; Abrusán 2011; Abeillé et al. 2020, a.o.).
