A Bot is Not a Polyglot: Designing Personalities for Multi-Lingual Conversational Agents

Conversational agents are becoming more common, influenced by the success of Siri and Alexa. As such, new methods and associated challenges of designing for conversational systems are emerging. One factor, unique to conversational agents, that we need to account for as designers, is personality. People attribute personalities to conversational agents, strongly influenced by social expectations, whether or not a particular personality was designed deliberately. In the case of multi-lingual agents, this creates additional challenges: direct translations don’t accommodate cultural variation. We discuss the design process and lessons learned from Radar Pace, a conversational coach that launched with support for five languages. We highlight successes and failures based on user study results and propose changes to the design process to avoid pitfalls for future agents.