BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Minimum Bayes-Risk Lattice Rescoring Methods for Statistical Machi
 ne Translation - Graeme Blackwood\, University of Cambridge
DTSTART:20110520T110000Z
DTEND:20110520T120000Z
UID:TALK31452@talks.cam.ac.uk
CONTACT:Thomas Lippincott
DESCRIPTION:Modern SMT systems incorporate multiple components\, statistic
 al models\, and processes. Translation is often factored as a series of mo
 dules with the output of one module serving as the input to the next. To a
 void propagation of errors\, it is better to avoid hard decisions by passi
 ng on as much information as possible to subsequent stages of the MT pipel
 ine\, usually in the form of a lattice or list of the most likely hypothes
 es. This enables the application of models that are difficult or impossibl
 e to apply in first-pass translation.\n\nI will describe several large-sca
 le SMT lattice rescoring procedures based on minimum Bayes-risk decoding\,
  starting with an efficient implementation of lattice MBR that uses weight
 ed path counting transducers to compute the required statistics. This impl
 ementation allows efficient generalisation of the MBR decoder to the task 
 of multiple-lattice system combination. I will conclude by describing a co
 nfidence-based lattice segmentation and MBR decoding framework\; this fram
 ework enables the targeted application of models intended to address parti
 cular deficiencies in SMT hypotheses.\n
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR