Proceedings of the Annual Meeting of the Cognitive Science Society

Listeners often strengthen the interpretation of expressions by deriving upper-bounded readings (URs), as when warm is understood as warm but not hot—an inference typically assumed to arise due to reasoning about a pragmatic scale 'warm, hot'. But, UR rates are also shown to vary substantially across and within scales. We investigate this diversity for gradable adjectives, aiming to explain why less URs arise for relative adjectives with relative scalemates ('warm, hot') than absolute ones ('likely, certain'). We tease apart a view that attributes this effect to differences in salience of the stronger term from one attributing it to the way standards are set differently for these adjective types. Two experiments manipulated alternative salience and comparison classes while controlling for adjective type, showing that comparison classes modulate UR rates 'rel, rel', but not 'rel, abs' scales, while alternative salience affects both. This pattern challenges salience-only accounts but is compatible with an account where URs depend on the way adjectival standards are derived from the comparison class.