Nile is a supervised, discriminative word alignment package that can make use of arbitrary and overlapping features. Our current supplied language-independent featureset enables accurate models of word alignment as tested on Arabic-English and Chinese-English language pairs. You can easily augment training with your own set of features specific to whatever language pair you are working with.
For details, see:
Feature-Rich Language-Independent Syntax-Based Alignment for Statistical Machine Translation.
(J. Riesa, A. Irvine, and D. Marcu). 2011. In Proceedings of EMNLP, pp. 497-507.
Hierarchical Search for Word Alignment
(J. Riesa and D. Marcu). 2010. In Proceedings of ACL, pp. 157-166.
Last modified: June 10, 2012.