Let's Chat! Can Virtual Agents learn how to have a Conversation?

Abstract

Intelligent virtual agents frequently engage the user in conversation. The underlying technology - often referred to as spoken dialogue systems - have experienced a revolution over the past decade, moving from being completely handcrafted to using data-driven machine learning methods. In this talk, I will review current developments including my work on using reinforcement learning and deep learning models, and evaluate these methods in the light of recent results from two large-scale studies: First, I will summarise results from a shared task, the End-to-End Natural Language Generation Challenge (E2E NLG) for presenting information in closed-domain task-based dialogue systems. Second, I will report our experience from experimenting with these models for generating responses in open-domain social dialogue as part of the Amazon Alexa Prize challenge. Throughout my talk, I will highlight challenges and opportunities of machine learning based response generation.