Improve the chatbot performance for the DB-CALL system using a hybrid method and a domain corpus

Abstract

This paper presents a chatbot for a Dialogue-Based Computer Assisted Language Learning (DB-CALL) system. The chatbot helps users learn language via free conversations. To improve the chatbot performance, this paper adopts a Neural Machine Translation (NMT) engine to combine with an existing search-based engine, and also extracts a small domain corpus for the topics of the DB-CALL system so that the chabot’s responses could be more related to the conversation topics. As a result of user evaluations, the performance of the chatbot was improved by using hybrid methods, achieving performance comparable to existing systems. The automatically extracted domain corpus has little help or even declines the chatbot performance as an auxiliary module of the DB-CALL system.