Doubling up: Two upper bounds for scalars

Most theories of scalar quantifiers, of whatever persuasion, assume a lexical lower-bound-only, 'at least' meaning for scalar quantifiers, offering pragmatic or grammatical mechanisms for deriving the upper bound (Carston 1990; Chierchia 2004; Horn 1972 and onwards). I have challenged the lower-bound analysis in Ariel (2004), proposing instead a circumbounded analysis for quantifier most, where the upper bound too is lexically specified. I here extend the analysis to some. The most important feature of the circumbounded analysis is that it splits into two what are commonly considered one and the same interpretation of 'less than all'. The first is a lexeme-level upper bound which asserts the speaker's commitment to the existence of some proper subset, the reference set, for which the predicate holds. The second is pragmatic, an exclusion of the complement set from the predication. Based on new questionnaire data, my main argument here is that even in contexts which clearly militate against a 'not all' interpretation, a nonmaximal upper bound is understood. More generally (and tentatively), I will cast doubt on the whole Gricean idea that linguistic semantics should be reduced to logic.