< back to Tutorials

EMPIRICAL METHODS FOR DIALOGUE SYSTEMS RESEARCH

Gregory Aist

This tutorial will present a comprehensive overview of empirical methods for building dialogue systems. We will cover phases of development including requirements gathering, design, implementation, testing, refinement, deployment, evaluation, customer support, and product maintenance. The emphasis throughout will be on how to turn design issues into empirical questions, and on learning a variety of techniques for answering those questions from data collected in the lab and in the field. We will spend about one-third of the time covering fundamental techniques, and the remainder on more advanced methods.

This tutorial will be timely for researchers with a background in a core area of NLP such as parsing who are planning to move into dialogue systems and wish to employ the same empirical rigor to dialogue as has become the norm in the parsing community. Those working in dialogue systems will benefit from an up-to-date overview of empirical methods across the wide range of development phases, particularly since methods for some phases such as testing, refinement, and support are less commonly known in the community. Students in NLP and in fields such as statistics or experimental design will be able to get a broad picture of the state of the art in applying empirical methods to dialogue systems. Finally, companies that are looking to launch a dialogue system effort will especially benefit from the requirements gathering and design sessions since we will present data-driven ways to figure out where in a business or a customer's interactions would substantially benefit from a dialogue systems application.

TUTORIAL OUTLINE

1. First session
   • Introduction
   • Goals and Requirements
   • Design and Implementation
   • Questions and Discussion
2. Second Session
   • Testing and Refinement
   • Deployment and Evaluation
   • Support and Maintenance
   • Questions and Discussion
   • Concluding Remarks

GREGORY AIST is a Research Associate at the University of Rochester. His scientific interests are in language, learning and computation. Language: human lexical learning, multi-topic conversation. Learning: effective strategies for instruction, language learning, the role of emotions in learning, learning procedural tasks. Computation: spoken language understanding, turn-taking, multimodal generation, intelligent tutoring systems, incrementality in spoken dialogue systems.


< back to Tutorials