Modelling context within a constraint-based account of quantifier usage
Recent work on numerically-quantified expressions has aimed to identify which components of their meaning are semantic and which are pragmatic. Pragmatically-oriented accounts assign a crucial role to contextual factors, such as the level of information requested in the preceding discourse and the availability of certain expressions to the speaker at the time of utterance. However, these models are typically imprecise as to which factors are relevant and how they interact. We discuss a recent proposal that treats numerical quantifier usage as a problem of multiple constraint satisfaction, and demonstrate its descriptive utility. In particular, we consider how the context-referring constraints within this model (governing prior mention of the numeral and the quantifier, granularity level, and informativeness) constitute a proposed definition of relevant context. In such a model, contextual factors can influence the output only if they are referred to in constraint definitions. Therefore, if a model of this type succeeds in capturing the meaning of use and numerical expressions, its constraint definitions specify which contextual factors are relevant to the choice of expression. We consider to what extent this account succeeds in modelling actual speaker behaviour, and whether it generalises to other domains of usage.