Better Alignments = Better Translations?

Automatic word alignment is a key step in training statistical machine translation systems. Despite much recent work on word alignment methods, alignment accuracy increases often produce little or no improvements in machine translation quality. In this work we analyze a recently proposed agreementconstrained EM algorithm for unsupervised alignment models. We attempt to tease apart the effects that this simple but effective modification has on alignment precision and recall trade-offs, and how rare and common words are affected across several language pairs. We propose and extensively evaluate a simple method for using alignment models to produce alignments better-suited for phrase-based MT systems, and show significant gains (as measured by BLEU score) in end-to-end translation systems for six languages pairs used in recent MT competitions. Comments Reprinted from: Better Alignments = Better Translations? Kuzman Ganchev, Joao Graca and Ben Taskar. In Proceedings of the 46th Annual Meeting of the Association of Computational Linguistics. Columbus, Ohio, June 16-17, 2008. This conference paper is available at ScholarlyCommons: http://repository.upenn.edu/grasp_papers/42 Better Alignments = Better Translations? Kuzman Ganchev Computer & Information Science University of Pennsylvania João V. Graça LF INESC-ID Lisboa, Portugal Ben Taskar Computer & Information Science University of Pennsylvania