Korean-English named entity alignment for machine translation of digital TV closed caption

We present a Korean to English machine translation system of closed captions for digital television. Named entities (NEs) convey important information of news articles and drama scripts. So, correct translation of NEs is indispensable for effective comprehension of TV programs. In order to achieve more accurate translation results, we adopted English and Korean NEs alignment into the base MT system. The resources for NEs and their translingual equivalences are acquired from the Web by an intelligent Web crawler. With these integrated approaches, we could raise the translation accuracy on average by 0.60 (mean opinion score) higher than the base MT system.