Framework to Specify Dialogues for Natural Interaction with Conversational Assistants Applied in Prompt Engineering

This article explores the significance of Natural Interaction (NI) in the context of conversational assistants (CAs), aiming to model user dialogues through written and/or spoken language. By NI notion, we consider it is a grammar-free communication language as close as possible to human interaction through written and spoken language, and serves as a highly intuitive means of interaction. It eliminates the need for users to learn new commands, making tasks more accessible. This research addresses the question of which elements and constraints can be employed in crafting NI dialogues for CAs. The emergence of large language models (as ChatGPT) has transformed dialogue specification, reducing the role of producers and enabling users to configure and qualify dialogues based on their needs. The response’s quality and user understanding now depend on how well dialogues are specified. This study introduces a framework, DINA, to assist dialogue specifiers, even those without technical or linguistic expertise in AI, CAs, or Human-Computer Interaction, in describing assistant guidelines, dialogue benefits, and other essential attributes for effective NI.