Modeling the Emergence of an Exemplar Verb in Construction Learning

Using a computational model of verb argument structure learning, we study a key assumption of the usage-based theory: that the acquisition of a construction relies heavily on the existence of a high-frequency exemplar verb that accounts for a large proportion of usages of that construction in the input. Importantly, unlike the psycholinguistic experiments that focus on the learning of an artificial novel construction using novel verbs, here we examine the acquisition of the English sentential complement construction from naturalistic input. Our results provide new insights into exemplar-based learning in the context of naturalistic input with multiple semantic classes, and a diverse set of constructions for the verbs.