Two-phase base noun phrase alignment in Chinese-English parallel corpora

A two-phase approach of automatically aligning bilingual base noun phrases from sentence-aligned Chinese-English parallel corpus is proposed in this paper. We conduct alignment in two phases: one deals with high-frequency base noun phrases by statistical co-occurrence information between parallel corpus, and the other deals with low-frequency base noun phrases using the bilingual lexical information and Dice coefficient similarity metrics. This can be reasonably considered to acquire higher recall without degrading the precision on the whole. Furthermore, our approach can escape from complex Chinese parsing problems and don't need to recognize Chinese base noun phrases accurately before the aligning process. Also, it can also relieve, to some extent, the serious impacts of error spread which may result from the unstable and impractical Chinese base noun phrases extraction tools. Another, dealing with high frequency noun phrases with statistical information also can realize the recognition of some non-compositional phrase pairs, which is difficult for pure syntax-based or lexicon-based systems to handle.