BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Improved Information Structure Analysis of Scientific Documents Th
 rough Discourse and Lexical Constraints - Yufan Guo\, University of Cambri
 dge
DTSTART:20130517T110000Z
DTEND:20130517T120000Z
UID:TALK45371@talks.cam.ac.uk
CONTACT:Ekaterina Kochmar
DESCRIPTION:Inferring the information structure of scientific documents is
  useful for\nmany down-stream applications.  Existing feature-based machin
 e learning approaches to this task require substantial training data and s
 uffer from limited performance.  Our idea is to guide feature-based models
  with declarative domain knowledge encoded as posterior distribution const
 raints.  We explore a rich set of discourse and lexical  constraints which
  we incorporate through the Generalized Expectation (GE) criterion.\nOur c
 onstrained model improves the performance of existing fully and\nlightly s
 upervised models. Even a fully unsupervised version of this model outperfo
 rms lightly supervised feature-based models\, showing that our approach ca
 n be useful even when no labeled data is available.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR