The strength of conservativity: evidence from learnability experiments

Abstract

One of the best-known semantic universals is determiner conservativity, roughly, the idea that the truth of sentences like most fish swim depends only on the things named by the determiner’s first argument (fish). Conservativity has been argued to reflect a fundamental property of grammatical architecture, a conclusion bolstered by evidence that adults (and children) can pair novel determiners with conservative meanings but not with minimally-different non-conservative meanings. Here we extend this work to ‘weakly’ conservative meanings (ones for which the truth of the quantificational sentence depends either on the determiner’s first argument or on its second argument). Such meanings are classically non-conservative but would be permitted under a weakened version of the generalization, which was designed to accommodate meanings like those of only and even. We find that adults cannot learn these kinds of novel determiner meanings either. This result suggests that the classical understanding of conservativity (on which ‘weakly’ conservative meanings are ruled out by virtue of being non-conservative) better describes the constraint that learners embody.