Dialogue Act Classification for Virtual Agents for Software Engineers during Debugging

A "dialogue act" is a written or spoken action during a conversation. Dialogue acts are usually only a few words long, and are often categorized by researchers into a relatively small set of dialogue act types, such as eliciting information, expressing an opinion, or making a greeting. Research interest into automatic classification of dialogue acts has grown recently due to the proliferation of Virtual Agents (VA) e.g. Siri, Cortana, Alexa. But unfortunately, the gains made into VA development in one domain are generally not applicable to other domains, since the composition of dialogue acts differs in different conversations. In this paper, we target the problem of dialogue act classification for a VA for software engineers repairing bugs. A problem in the SE domain is that very little sample data exists - the only public dataset is a recently-released Wizard of Oz study with 30 conversations. Therefore, we present a transfer-learning technique to learn on a much larger dataset for general business conversations, and apply the knowledge to the SE dataset. In an experiment, we observe between 8% and 20% improvement over two key baselines.