Zahra Mirrazi (UMass, Amherst) & Hedde Zeijlstra (Göttingen)
A non-lexical approach to NEG-RAISING
Pragma-semantic approaches to Neg-Raising (NR) take NR readings to be the result of an excluded middle inference, either in terms of a presupposition, or in terms of scalar implicatures, which is special to a certain group of predicates known as Neg-Raising Predicates (NRPs) like 'think'. While successful in accounting for many aspects of NRPs, these approaches face some non-trivial problems. (i) There are contexts under which NRPs receive a non-NR reading without resulting in a presupposition failure. (ii) Some non-NRPs (e.g. non-factive 'know') can get a NR reading.We propose a new implementation of a scalar implicature account to NR. Our analysis has two components: duality and strengthening of subdomain alternatives. We take the basic reading of negated NRPs to involve existential quantification over worlds where not-p holds, as a result of equivalence with the basic meaning of negated NRPs which involves a negated universal quantifier over worlds where p holds (duality). Parallel to contemporary implicature accounts of Free Choice and Homogeneity, this existential reading can be strengthened to a universal one via application of an exhaustification operator. Under this view, the (un)availability of NR readings for duality-allowing modals is reduced to whether exhaustification applies to the whole set of subdomain alternatives (yielding the strengthened reading) or over a subset after pruning singleton sets (yielding the weak reading).
We take (i)-(ii) to show that the ability to trigger a NR reading is not a lexical property of NRPs. Our approach to NR is the only approach that can account for this. All other theories of NR take NRPs to carry some unique lexically-encoded property. Since the application of exhaustification is context-dependent, we allow every negated universal modal whose presuppositions do not block duality, to be able to yield a NR-reading, provided that the whole set of subdomain alternatives is contextually relevant.