A Discriminative Model for Mongolian-English Word Alignment

In this paper we have developed a discriminative word alignment method using the liner CRF model for Mongolian-English language pair. Discriminative approaches have the ability to use large variety of features flexibly and to combine information from various sources. According to the differences between Mongolian and English, we discuss and select the different features, and experiments on Mongolian-English task demonstrate our word alignment model's superiority in performance over the popular GIZA++.