Toward Lexical Acquisition during Dialogues through Implicit Confirmation for Closed-Domain Chatbots

This paper proposes a lexical acquisition framework for a closeddomain chatbot. It learns the ontological categories of unknown terms in dialogues through implicit confirmation instead of using explicit questions which are too abrupt. Analysis of human participant responses to implicit confirmation requests confirmed the feasibility of our approach: the response expressions differed in a manner that indicated the correctness of an unknown term’s predicted category. We thus propose a two-tiered method to predict unknown term categories that attempts to predict the most specific category possible and backs off to a more general category when the predicted category is not sufficiently confident. Direct evaluation showed that this two-tiered method makes correct category predictions 54.3% more often than previously proposed explicit confirmation method.