How many manys? Exploring semantic theories with data-driven computational models

Abstract

We use a data-driven computational inference approach to address the question whether it is plausible to maintain that there is a stable core semantics that governs the interpretation of cardinal and proportional many across different contexts. Adopting the idea that the denotation of many is a function of a stable threshold parameter that applies to a contextually-variable probability distribution that captures prior expectations, we demonstrate that it is possible to maintain that there is a single fixed threshold for many’s cardinal and proportional use, although models that allow for non-uniform thresholds or lexical ambiguity may have a slightly better empirical fit to our data.