Abstract
We examine the acquisition of subcategories of Gradable Adjectives (GAs). We first show that robust patterns of adverbial modification in natural language sort GAs according to scalar structure: proportional modifiers (e.g., completely) tend to modify absolute maximum standard GAs (e.g., full), while intensifiers (e.g., very) tend to modify relative GAs (e.g., big). We then show in a word-learning experiment that 30-month-olds appear to be aware of such distributional differences and recruit them in word learning, assigning an interpretation to a novel adjective based on its modifier. We argue that children track both the range of adjectives modified by a given adverb and the range of adverbs modifying a given adjective, and use such surface-level information to classify new words according to possible pre-existing semantic representations.