Google Tech Talks
October 22, 2008
In this talk I'll outline our work at the University of Edinburgh to model machine translation (MT) as a probabilistic machine learning problem. Although MT systems have made large gains in translation quality in recent years, most current approaches are based on a hand engineered pipeline of disparate models linked by heuristics. I'll motivate why MT provides an interesting, but hard, structured learning problem, and describe our recent work tackling it with both discriminative linear models and generative Bayesian models. In doing so I'll demonstrate how powerful tools from machine learning, such as high dimensional sparse feature functions, regularisation and hierarchical Dirichlet priors, can be effectively applied to modelling the translation process.
Speaker: Phil Blunsom
Phil Blunsom is a Research Fellow in the School of Informatics at the University of Edinburgh. He is part of the Machine Translation research group. Currently he is working on the application of machine learning techniques to machine translation with Miles Osborne.
Phil completed my PhD at the University of Melbourne, Australia, under the supervision of Timothy Baldwin, Steven Bird and James Curran.
My thesis research focused on the application of log-linear graphical models, such as conditional random fields, to complex natural language processing tasks. In particular, how discriminative models can improve areas such as machine translation and multilingual lexical acquisition.