Department Seminar - by Dr. Yang Liu (UTDallas)

Updated on Thu, 02/04/2010 - 4:27pm

Details: Processing "natural" language: recent research on text messages and meeting recordings

Speaker: Dr. Yang Liu
When: Fri, 02/26/2010 - 11:08am - 12:08pm
Room: CH 430
Abstract: Natural language processing research has seen significant progress in the past decades. Compared to the efforts on well written text (e.g., news articles), there are fewer studies using less formal language (such as blog, forum, talk shows, text messages, etc.). Recently there has been increasing interest in these genres in language processing community. In this talk, I will present some results from our research on dealing with text messages and multiparty meetings.

Specifically, for text messages, I will discuss text normalization issues. We use a statistical classifier to model the abbreviation phenomenon, and generate multiple abbreviation hypotheses for a word. A reverse mapping table is used to enable prediction of English words from the abbreviation, using reranked scores based on different knowledge sources. For meeting recordings, I will focus on the task of automatic summarization. We explore various unsupervised and supervised approaches for summarization. For supervised learning, we use sampling and regression to address the imbalanced data issue and low human agreement problem. In addition, we also use multiple speech recognition hypotheses to cope with recognition errors. These yield improved summarization performance.
Biography: Dr. Yang Liu has been an Assistant Professor in Computer Science Department at the University of Texas at Dallas since 2005.
She received her B.S and M.S degree from Tsinghua University in China, and Ph.D from Purdue University. She was a researcher at International Computer Science Institute (ICSI) in Berkeley before she joined UTD. Her research interest is in speech and language processing. She received the NSF CAREER award in 2009, and the Air Force Young Investigator Research Program award in 2010.