Annotated Corpora for Word Alignment between Japanese and English and its Evaluation with MAP-based Word Aligner

This paper presents two annotated corpora for word alignment between Japanese and English. We annotated on top of the IWSLT-2006 and the NTCIR-8 corpora. The IWSLT-2006 corpus is in the domain of travel conversation while the NTCIR-8 corpus is in the domain of patent. We annotated the first 500 sentence pairs from the IWSLT-2006 corpus and the first 100 sentence pairs from the NTCIR-8 corpus. After mentioned the annotation guideline, we present two evaluation algorithms how to use such hand-annotated corpora: although one is a well-known algorithm for word alignment researchers, one is novel which intends to evaluate a MAP-based word aligner of Okita et al. (2010b).