The Fauconnier-Ladusaw analysis of negative polarity licensing (that NPIs are licensed in the scope of downward entailing operators) continues to be the benchmark theory of negative polarity. In this paper, I consider some of the moves that are needed to maintain its basic intuition in some recalcitrant arenas: negative polarity licensing by only, adversatives, superlatives, and conditionals. We will see that one has to (i) use a notion of entailment that I call Strawson Entailment, which deals with presuppositions in a particular way, and (ii) prohibit (even natural) context change during an inference. The purpose of this paper is to investigate how to justify these constraints and to see in detail how the semantics of the problematic constructions has to work in order for these moves to successfully rescue the Fauconnier-Ladusaw analysis. I will first show the two assumptions at work in the analysis of NPI licensing by only and adversatives (building on proposals by Kadmon & Landman). I then turn to NPI licensing in the antecedent of conditionals. The standard Stalnaker-Lewis semantics for conditionals-ifp, q is true iff q is true in the closest p-woild(s)-might make one suspect that once one has an explanation for NPI licensing by superlatives, that would immediately deliver an explanation for NPI licensing in conditionals. But it turns out that the particular analysis that seems appropriate for NPI licensing by superlatives cannot plausibly carry over to conditionals. Instead, one does better by appealing to an alternative analysis of conditionals, one that I have elsewhere argued for on independent grounds. I BASICS Negative polarity items (NPIs) are expressions that can prototypically occur in the semantic scope of negation but not in 'positive' environments. Two examples are arty and ever. (1) a. I don't think we have any potatoes. #1 think we have any. potatoes, b. I don't think there will ever be another Aristotle. #1 think there will ever be another Aristotle. Since NPIs occur not just in the scope of negation but also in a variety of other 'affective' environments (the term is due to Klima 1964), one needs a theory of NPI licensing. Building on an idea from Fauconnier (1975, 1979),

No abstract

No abstract

In recent work, MacFarlane [] and Egan et al. [] have used new observations about epistemic modals to motivate radical new semantic analyses. MacFarlane introduces a triple-indexed semantics which makes sentences with epistemic modals not just sensitive to the context of utterance and an index of evaluation but also to a context of assessment. Egan et al. propose that epistemically modalized sentences do not express standard propositions but functions from individuals to propositions, something like centered propositions. We will argue that these moves are not only problematic but also unnecessary.

Epistemic modals are standardly taken to be context-dependent quantifiers over possibilities. Thus sentences containing them get truth-values with respect to both a context and an index. But some insist that this relativization is not relative enough: `might'-claims, they say, only get truth-values with respect to contexts, indices, and—the new wrinkle—points of assessment (hence, cia). Here we argue against such “relativist” semantics. We begin with a sketch of the motivation for such theories and a generic formulation of them. Then we catalogue central problems that any such theory faces. We end by outlining an alternative story.