2-step word alignment framework for Thai-English statistical machine translation

• Main Authors: Prasert Luekhong, Taneth Ruangrajitpakorn, Rattasit Sukhahuta, Thepchai Supnithi
• Format: Journal
• Published: 2018
• Subjects: Computer Science
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85040814274&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/57064
• Institution: Chiang Mai University

© 2017 International Information Institute. This paper presents a framework of a new word alignment process for SMT and the translation table improvement method with bilingual dictionary that was the lexical probabilistic tuning methodology for translation table in SMT. First, the alignment method was designed to include the quality of using dictionary as prior knowledge and the ability of co-occurrence to fill unknown words. By testing the proposed framework against the renowned GIZA, we applied an alignment model from both systems to Moses for proving its usefulness in a practical hierarchical phrase-based translation usage and exploited a BLEU score as a measurement. The case study in this work focused on Thai to English translation. The testing results showed the proposed method can overrun the result of GIZA IBM model-4 by 2.09 BLEU points.