As one of semi-supervised learning approach, cross-lingual projection leverages existing resources from a resource-rich language when building tools for resource-poor languages. In this paper we attempt to make use of word embedding with anchor based label propagation to improve the accuracy of a cross-lingual projection task: cross-lingual part-of-speech tagging under the graph-based framework. Our approach uses bilingual parallel corpora and labeled data from the resource-rich side assuming that there is no labeled data or only a few labeled data in resource-poor language. The results suggest the efficacy of our approach over traditional label propagation with lexical feature for projecting part-of-speech information across languages, and show that a few of labeled data help to enhance the effect a lot in cross-lingual task.
Benzer Makaleler | Yazar | # |
---|
Makale | Yazar | # |
---|