Effects of interface interactivity on collecting language data to power dialogue agents

Abstract

Conversation is one of the easiest modes of communication. Interactive dialogue agents are promising natural interfaces for people to interact with machines. The building of these agents, however, suffers from lacking quality language data for supporting the generation of conversation-like system responses. In this paper, we explore using an interactive chat bot to elicit more naturalistic language data from Chinese-speaking and English-speaking workers. We present two studies to examine the impact of interface interactivity on data quality as well as the ultimate experience of dialogue agent users. Results show that online workers preferred working with the interactive chat bot than a static questionnaire. Users speaking different languages also have different perceptions on the enjoyability of the dialogue agent powered by our data collection method.