Alignment link projection using transformation-based learning

Abstract

We present a new word-alignment approach that learns errors made by existing word alignment systems and corrects them. By adapting transformation based learning to the problem of word alignment, we project new alignment links from already existing links, using features such as POS tags. We show that our alignment link projection approach yields a significantly lower alignment error rate than that of the best performing alignment system (22.6% relative reduction on English- Spanish data and 23.2% relative reduction on English-Chinese data). © 2005 Association for Computational Linguistics.