carp@colloquial.com
Title:LingPipe: Architecture and Algorithms
Time:Thursday October 30, 11:30 - 12:30
Place:CS Conference Room in MUDD
Abstract:
I will discuss the architecture, algorithms and performance
of our open-source natural language software, LingPipe.
LingPipe provides trainable named-entity, sentence detection
and within-document coreference. I'll detail the statistical
models, data structures, decoder, and I/O integration
that support training and testing on multilingual data.
Performance of version 1.0 is 100K words/second for
named-entity detection with an F score of .93 on MUC7 data
trained without any gazetteers or other dictionaries.
More details, including a download link, will be available
from 23 October 2003 at:
Home Page: http://www.alias-i.com/lingpipe
About the speaker: Before joining Alias I, Bob Carpenter had been at SpeechWorks International, and the Lucent Technologies Multimedia Communications Laboratory. Prior to Lucent, he was an Associate Professor of Computational Linguistics in the Philosophy Department at Carnegie Mellon. His Ph.D. is from the University of Edinburgh. For more details, see Bob Carpenter's Projects.