[Colloquium] JOB TALK: Steven Bethard, UCBoulder on February 3

[Katie Casey]

DEPARTMENT OF COMPUTER SCIENCE

UNIVERSITY OF CHICAGO

Date: Friday, February 3, 2012
Time: 2:30 p.m.
Place: RY 251

----------------------------------------------------------

Speaker:		Steven Bethard

From:		University of Colorado, Boulder

Web page:	http://clear.colorado.edu/~bethard/index.html

Title: 		Extracting Timelines from Unstructured Text

Abstract: Finding the events in a narrative and placing them in order along a timeline is a crucial task for applications like analyzing clinical records for disease patterns or generating reading comprehension questions to test children's knowledge of timelines in stories. However, most natural language narratives do not provide event timelines in structured form. Rather, they must be inferred from phrases like "The medication was discontinued 24-hours ago", where the "medication" event ends one day before the document was written, or from phrases like "Before the bear observed him, the traveller climbed a tree", where the "observed" event occurs after the the "climbed" event.

In this talk, I will first present a common architecture for timeline extraction: a series of supervised machine learning components trained on example texts whose timelines have been annotated manually by humans. Then I will show how to improve this approach by leveraging additional plain text data that has not been annotated by humans but nonetheless reveals some patterns in how humans talk about time. Finally, I will present an alternative approach to timeline extraction, motivated by linguistic and psychological theories, that produces more complete timelines in the form of dependency structures.

Host: John Goldsmith

Refreshments will be served in Ryerson 255 following the talk at 3:30.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20120120/22a32d2e/attachment.htm