Proceedings of the Annual Meeting of the Cognitive Science Society

A hallmark of natural language is the innovative reuse of ex-isting words. We examine how adjectives extend over timeto describe nouns and form previously unattested adjective-noun pairings. Our approach is based on the idea of chainingthat postulates word meaning to extend by linking novel ref-erents to existing ones that are close in semantic space. Wetest this proposal by exploring a set of models that learn toinfer adjective-noun pairings from historical text corpora fora period of 150 years. Our findings across three diverse setsof adjectives support a chaining mechanism that is sensitiveto semantic neighbourhood density, best captured by an exem-plar model of category extension. This work sheds light on thegenerative cognitive mechanisms of word usage extension.