Repair in a task-oriented chatbot, and the question of communication
skills for AI
Abstract
Repair describes the process through which participants in conversation
address problems in speaking, understanding, and hearing. In
interactions with AI-driven chatbots, repair helps users clarify their
intents and addresses problems in understanding intent experienced by
chatbots. This paper represents the first attempt to describe repair
strategies in a task-oriented text-based chatbot from a user-centred
perspective. It is based on the analysis of simulated user interactions
with a chatbot facilitating health appointment bookings. The analysis
shows that the self-repair strategies which users draw on most
frequently (e.g. rephrasing) are not necessarily the ones which prompt
the bot to provide relevant responses, whereas more successful
self-repair strategies (e.g. restating the intent) tend to be more
opaque to users and thus used relatively infrequently. This suggest that
the notion of communicative competence needs to be re-thought for
conversational AI and that communication skills for AI need to be taught
explicitly.