CLIP Researchers Active at NAACL Conference

Faculty and graduate students associated with the Computational Linguistics and Information Processing (CLIP) Laboratory were active at the North American Chapter of the Association for Computational Linguistics conference (NAACL) held earlier this month in Denver, Colorado.

NAACL is considered the top conference in natural language processing, drawing researchers from across the U.S. and beyond for a series of workshops, tutorials and networking events.

Philip Resnik, a professor of linguistics with an appointment in UMIACS, was a co-chair on the NAACL program committee for papers in the area of “NLP [natural language processing] for Web, Social Media and Social Sciences.”

Resnik also presented his research on computational modeling to detect depression in social media users. The paper, co-authored with UMD computer science doctoral students William Armstrong, Leo Claudino, Thang Nguyen, and Viet-An Nguyen, included input from former UMIACS member Jordan Boyd-Graber, now an assistant professor of computer science at the University of Colorado, Boulder.

Hal Daumé, an associate professor of computer science and CLIP Lab director, along with Microsoft Research Senior Research Scientist John Langford, gave a tutorial titled, “Hands-on Learning to Search for Structured Prediction.”

The goal of the tutorial, Daumé says, was to give participants experience writing small programs to perform structured prediction for a variety of tasks, such as sequence labeling and dependency parsing.

He adds that structured prediction is like translating Chinese to English. Instead of making a single decision on how to translate the text word-for-word, it’s necessary to look at mapping out an entire English sentence as a whole.

“Problems like this, in which we must make coherent joint predictions over a collection of variables, can be solved using learning-to-search techniques,” says Daumé, who also has an appointment in UMIACS. “We presented recent advances in this technology, as well as provided participants an opportunity to use this technology in a hands-on way to solve problems of their own choosing, akin to a ‘flipped classroom.’”

UMD computer science doctoral students He He and Sudha Rao and University of Illinois Urbana-Champaign doctoral student Kai-Wei Chang assisted with the tutorial led by Daumé.