Extraction of phrases useful for machine translation

To improve the quality of a machine translation system, it is useful to extract and store words and phrases which are salient in a domain and should be translated in a specific way. We tried several existing technologies for salient word extraction and compared their mechanisms and quality. Based on the comparison, we propose a new method which includes pre-processing of sentences in corpora and utilization of syntactic information. We evaluated the usefulness of our new method.