Recent years have seen strong improvements in language technologies such as spoken language understanding, machine reading comprehension, and question answering. A major beneficiary of these technologies are so-called dialogue systems, such as virtual personal assistants. While classically many of these systems were hand-crafted, machine learning components are essential today. This talk provides a high-level overview over machine learning approaches in dialogue systems. We introduce the main modules of the classical pipeline and how they can be replaced by data-driven models -- usually deep neural networks. Finally, we discuss open challenges posed by data and label scarcity, as well as inadequate automated metrics.
Hannes Schulz received a BSc in cognitive science from the university of Osnabrück and an MSc in computer science from the university of Freiburg. During his PhD in Bonn, Hannes worked on deep neural network technologies for image segmentation. Today, Hannes is a researcher at Microsoft Research Montreal, with a focus on dialogue systems.
Microsoft Research was founded in 1991, with the aim of doing fundamental computer research. Research at Microsoft today encompasses the Microsoft Research organization along with researchers and scientists who are embedded across the company. Researchers and scientists at Microsoft collaborate with Microsoft product and service teams, and with researchers and scientist from academic institutions and industrial researchers. The Montreal lab was founded in 2017. Engineers and researchers in the lab work on natural language processing, dialogue, reinforcement learning, and fairness and transparency.