Reproducible Research and the Common Task Method

The ‘Reproducible Research’ idea posits that publishing data and code, not just statistical summaries, makes for better and faster science. In particular, shared datasets and shared evaluation metrics lower barriers to entry, and allow meaningful comparison of scientific hypotheses with engineering algorithms.

In this lecture, Mark Liberman will describe the origins and development of the ‘Common Task’ method in DARPA’s human language technology program, its broader influence on recent research and development practices, and its lessons for the future. Large, shared datasets and well-defined evaluation metrics allow the steady improvement of technologies a decade or more in advance of commercial viability. There are important opportunities to apply similar ideas in a wide variety of areas, from autism research to STEM education and writing instruction.

Registration information coming soon.
If this lecture is videotaped, it will be posted here after production.