Proceedings of the Annual Meeting of the Cognitive Science Society

Language is inherently flexible: people continually generalize over observed data to produce creative linguistic expressions. This process is constrained by a wide range of factors, whose interaction is not fully understood. We present a novel study of the creative use of verb constructions ``in the wild'', in a very large social media corpus. Our first experiment confirms on this large-scale data the important interaction of category variability and item similarity within creative extensions in actual language use. Our second experiment confirms the novel hypothesis that low-frequency exemplars may play a role in generalization by signaling the area of semantic space where creative coinages occur.