BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:NLIP reading group: Fast and Robust Joint Models for Biomedical Ev
 ent Extraction - Andreas Vlachos (University of Cambridge)
DTSTART:20111020T110000Z
DTEND:20111020T120000Z
UID:TALK33717@talks.cam.ac.uk
CONTACT:Jimme Jardine
DESCRIPTION:Andreas will be taking us through the following:\nhttp://www.a
 clweb.org/anthology/D/D11/D11-1001.pdf\n\nExtracting biomedical events fro
 m literature\nhas attracted much recent attention. The bestperforming\nsys
 tems so far have been pipelines\nof simple subtask-specific local classifi
 ers. A\nnatural drawback of such approaches are cascading\nerrors introduc
 ed in early stages of the\npipeline. We present three joint models of\ninc
 reasing complexity designed to overcome\nthis problem. The first model per
 forms joint\ntrigger and argument extraction\, and lends itself\nto a simp
 le\, efficient and exact inference\nalgorithm. The second model captures\n
 correlations between events\, while the third\nmodel ensures consistency b
 etween arguments\nof the same event. Inference in these models\nis kept tr
 actable through dual decomposition.\nThe first two models outperform the p
 revious\nbest joint approaches and are very competitive\nwith respect to t
 he current state-of-theart.\nThe third model yields the best results repor
 ted\nso far on the BioNLP 2009 shared task\,\nthe BioNLP 2011 Genia task a
 nd the BioNLP\n2011 Infectious Diseases task.\n\n\n\n@inproceedings{riedel
 2011fast\,\n  title={Fast and Robust Joint Models for Biomedical Event Ext
 raction}\,\n  author={Riedel\, S. and McCallum\, A.}\,\n  booktitle={Proce
 edings of the Conference on Empirical methods in natural language processi
 ng (EMNLP’11)}\,\n  year={2011}\n}\n\n
LOCATION:GS15\, Computer Laboratory
END:VEVENT
END:VCALENDAR