Abstract:

Much recent work has postulated lexical entries for such diverse items as gradable adjectives (e.g., ‘tall’, ‘big’), vague quantifiers (e.g., ‘many’, ‘few’), and probability expressions (e.g., ‘probably’, ‘likely’) in terms of a comparison with a contextually supplied threshold variable. We propose here a pragmatic account of the resolution of these elusive threshold values as a function of interlocutors’ contextually-variable prior expectations. More concretely, we introduce a model for computing thresholds from prior expectations that are (approximately) optimal under general pragmatic pressure towards informativity and generality. Despite its high level of abstraction, the model appears to be a surprisingly good predictor of experimental data on the use and interpretation of vague expressions.