Supervisor

Suitable for

Abstract

Baysian models of word alignment for Machine Translation (medium hard) A key step in building Machine Translation (MT) models is word alignment. Given a parallel corpus of sentences and their translation we need to automatically discover which words in the input language correspond to words in the output language. Current models use very limited context and only consider words in isolation, the aim of this project will be to use ideas from Bayesian generative models to extend the conditioning context and learn better word alignments.