English-Korean Named Entity Transliteration Using Substring Alignment and Re-ranking Methods

摘要

In this paper, we describe our approach to English-to-Korean transliteration task in NEWS 2012. Our system mainly consists of two components: an letter-to-phoneme alignment with m2m-aligner,and transliteration training model DirecTL-p. We construct different parameter settings to train several transliteration models. Then, we use two re-ranking methods to select the best transliteration among the prediction results from the different models. One re-ranking method is based on the co-occurrence of the transliteration pair in the web corpora. The other one is the JLIS-Reranking method which is based on the features from the alignment results. Our standard and non-standard runs achieves 0.398 and 0.458 in top-1 accuracy in the generation task.