I can't believe it's not lexical: Deriving distributed veridicality

Abstract

Given the assumption that selection is a strictly local relationship between a head and its complement, we expect the ability of a head to take a particular argument to be insensitive to linguistic material above that head. The verb believe poses a puzzle under this view: while believe ordinarily only permits declarative clausal complements, interrogative complements are allowed when believe occurs under clausal negation and can or will, and a veridical reading becomes available. I argue that this provides evidence that believe is not simply a standard Hintikkan representational belief verb, but rather is fundamentally question-embedding,and that the verb's lexical semantics, including an excluded middle presupposition, interact with the modal and negation to derive the veridicality of can't believe. I conclude that veridicality need not be lexical: the right mix of semantic ingredients can conspire to yield a veridical interpretation, even if those ingredients are distributed across multiple lexical items.