Continual Learning of Conversational Skills

In previous chapters, we have discussed how a chatbot can learn world knowledge (e.g., entities, facts, concepts) to generate more relevant responses and answer user questions (in Chap. 3), how it can improve its quality of response and avoid going out of context (in Chap. 5) and how it can acquire knowledge during conversation to understand user utterance better and serve user better than before (in Chaps. 4 and 6). All these qualities are essential to building a successful chatbot system that can respond satisfyingly and perform tasks well on users’ behalf. However, another important aspect that it should possess is the qualities of sensitivity, self-awareness and understanding of users’ (interlocutors) characteristics in order to best model its responses. These qualities are what separate us humans from machines. Specifically, the chatbot needs to learn users’ behaviors, preferences, emotions, moods, opinions and situations and leverage these pieces of knowledge while crafting its responses. This chapter focuses on continual learning of conversational skills.