An EMD Based Approach to Transliteration Unit Alignment between English and Chinese

Machine transliteration is to translate the proper nouns in the source language according to its pronunciation into the target language. Recent orthographical based approach has improved the performance of machine translation significantly. Focusing on the transliteration unit alignment that provides a fundamental parameter for the model, this paper adopts a semi-supervised EMD approach-applying discriminative model over the original EM results. The experiment results prove it is substantial to the improvement of the transliteration performance.