Leveraging the Crowd to Support the Conversation Design Process

Abstract

Building a chatbot with human-like conversation capabilities is essential for users to feel more natural in task completion. Many designers try to collect human conversation data and apply them into a chatbot conversation, aiming that it could work like a human conversation. To support conversation design, we propose the idea of inviting the crowd into the design process, where crowd workers contribute to improving the designed conversation. To explore this idea, we developed ProtoChat, a prototype system that supports a conversation design process by (1) allowing the crowd to actively suggest new utterances based on designers' pre-written design and (2) visually representing crowdsourced conversation data so that designers can analyze and improve their conversation design. Results of an exploratory study indicated that the crowd is helpful in providing insights and ideas as designers explore the design space.