BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Inducing Synchronous Grammars for Machine Translation - Phil Bluns
 om\, University of Oxford
DTSTART:20091127T120000Z
DTEND:20091127T130000Z
UID:TALK20473@talks.cam.ac.uk
CONTACT:Laura Rimell
DESCRIPTION:In this talk I'll outline my work modelling statistical machin
 e translation\n(SMT) as a probabilistic machine learning problem.\nAlthoug
 h SMT systems have made large gains in translation quality in recent\nyear
 s\, most are currently induced using a hand engineered  pipeline of\ndispa
 rate models linked by heuristics. Although such  techniques are effective\
 nfor translating between related languages  (e.g. English and French)\, th
 ey fail\nto capture the latent structure necessary to translate between la
 nguages which\ndiverge significantly  in syntactic structure\, such as Chi
 nese and English.\nI'll present non-parametric Bayesian models for inducin
 g synchronous context\nfree grammars. These models are capable of learning
  the latent structure of\ntranslation equivalence from a corpus of paralle
 l string pairs. I'll discuss\nthe difficult inference problems posed by su
 ch models and describe Monte Carlo\nsampling techniques that can help solv
 e them. Finally I'll present experiments\ndemonstrating competitive result
 s on full scale translation evaluations.
LOCATION:SW01\, Computer Laboratory
END:VEVENT
END:VCALENDAR