Still Phrase-Based and Proud of it: NRC's Participation in NIST OpenMT 2012

George Foster (george (point) foster <at> nrc (point) gc (point) ca)

CNRC - Gatineau

Le 14 novembre 2012 à 11 h 30

Salle 3195, Pavillon André-Aisenstadt


NIST OpenMT is the longest-running series of evaluations of Machine Translation technology, and one of the most important, attracting participation from many top MT groups. In 2012, NRC's Portage entry did extremely well in this evaluation (http://www.nist.gov/itl/iad/mig/openmt12results.cfm). This result was quite surprising, since Portage is a phrase-based system with no linguistic syntax, which is thought to be advantageous for language pairs such as Chinese-English. In this talk, I will describe the Portage NIST system, focussing on four techniques that were key to its success: batch MIRA tuning, hierarchical reordering, diverse phrase extraction, and domain adaptation using mixture models.


Pour recevoir les annonces hebdomadaires par courriel, visitez http://rali.iro.umontreal.ca/rali/?q=fr/node/1631

Liste des autres séminaires