TIntuitive Information Access: Answering Questions Using the Web, Databases, and VideoT

Jimmy Lin

Massachusetts Institute of Technology

UMIACS Computational Linguistics Colloquium

February 10, 2004, 2:00pm, AVW Room 2460

Intuitive Information Access: Answering Questions using the Web, databases, and video Question answering, which lies at the intersection between natural language processing and information retrieval, has recently emerged as a potential solution for overcoming the information overload problem faced by today's users.  The goal is to develop systems that understand information requests phrased in everyday language and respond with succinct answers.  I will describe various questions answering systems that I have developed over the past few years: Aranea, a Web-based QA system that integrates semi-structured database techniques with redundancy-based statistical methods.  The system has performed consistently well at several TREC QA evaluations. 

Naturally, not all information is textual: the application of question answering technology to other modalities such as video presents both interesting challenges and exciting opportunities.  I will describe Spot, a video-based question answering prototype that extracts high-level event semantics from surveillance footage and integrates visual and linguistic knowledge under shared representations.  Finally, I will describe experiments with extracting and indexing syntactic relations from text and performing question answering at the level of "key relations" in addition to keywords. Deeper analysis of documents leads to greater precision in answering natural language questions.

 About the Speaker:

Jimmy Lin is a Ph.D. student at MIT CSAIL working with Boris Katz on question answering and Alec Marantz on event structure and the syntactic encoding of predicate argument structures.  For the past several years, he has been working on many different aspects of question answering, both at MIT and Microsoft Research.  Recently, he has been working on event representations, with the ultimate hope of integrating computational and theoretical approaches to linguistics.

 For the colloquium series schedule, see the UMD Computational

http://www.umiacs.umd.edu/research/CLIP/colloq/ If you are interested in meeting with the speaker, please contact Doug <TUhttp://www.glue.umd.edu/~oard/UT>  Oard (oard@umiacs.umd.edu

<TUmailto:oard@umiacs.umd.eduUT>).