This paper has two goals. First, it defends my circumbounded analysis for scalar quantifiers most (Ariel, 2004) and some, according to which the upper bound is lexically specified. Based on new questionnaire results, I here argue against discourse-based counter-arguments against the circumbounded analysis. At the same time, the paper also makes a more general, methodological point about the (mis)use of discourse patterns in linguistic analyses. I argue that discourse patterns directly reflect only discourse/pragmatic principles. Hence, we cannot simply argue for or against specific semantic analyses based on prevalent discourse patterns, for the correct account for the pattern may be pragmatic, rather than semantic.
See Ariel (2012) about the two concepts of upper bound one linguistic and one pragmatic needed for scalars.
Horn (2009) also cites psycholinguistic experiments as supporting his analysis (e.g. Papafragou and Musolino 2003) but see Ariel (2004 2006 2012) for counter-arguments and reinterpretation of these psycholinguistic data. Neither Horn nor others I should add have offered an explanation for my questionnaire data which argue against the lower-bound semantic analysis and strongly support the lexical circumbounded analysis for most.
In Ariel (2012) I show that the clearly circumbounded part of patterns with the supposedly lower-bound-only some as well.