Creating Dialogues using Argumentation and Social Practices

Abstract

Research in chatbots is already more than fifty years old, starting with the famous Eliza example. Although current chatbots might perform better, overall, than Eliza the basic principles used have not evolved that much. Recent advances are made through the use of massive learning on huge amounts of resources available through Internet dialogues. However, in most domains these huge corpora are not available. Another limitation is that most research is done on chatbots that are used for focused task driven dialogues. This context gives a natural focus for the dialogue and facilitates the use of simple reactive rules or frame-based approaches. In this paper, we argue that if chatbots are used in more general domains we have to make use of different types of knowledge to successfully guide the chatbot through the dialogue. We propose the use of argumentation theory and social practices as two general applicable sources of knowledge to guide conversations.