Automatic extraction of entries for a machine translation dictionary using bitexts
Julia Aymerich, Hermes Camelo · 2007
It is a well-known fact that Machine Translation (MT) systems greatly benefit from user feedback and constant enhancement of the system’s dictionaries. However, getting translators to provide feedback in a consistent and timely fashion is difficult given the time constraints under which translators work. Without feedback from translators, the effort spent on postediting is lost for future translations. This paper presents a method recently incorporated in the PAHOMTS ® systems to extract truly useful dictionary suggestions using bitexts of past translations in combination with the PAHOMTS ® engine, and without any translator involvement. The suggestions are restricted to a few types of common entries and are ranked in order of priority depending on their frequency of occurrence in PAHO’s bilingual corpus. While dictionary entries are suggested automatically by the extractor, a human operator must validate the entries. The process of extraction and incorporation of dictionary entries is described in detail. The setting PAHOMTS ® is a well-established rule-based transfer MT system that has been operational at the Pan American Health Organization (PAHO) for over 25 years (Vasconcellos and León, 1988). It currently translates in six language directions (English-Spanish, Spanish-