Multi-Engine Machine Translation
- LecturerMr. Wei-Yun Ma (Ph.D. candidate in Computer Science at Columbia University)
Host: Keh-Jiann Chen - Time2014-01-16 (Thu.) 10:30 ~ 12:00
- LocationAuditorium 106 at new IIS Building
Abstract
The task of Multi-Engine Machine Translation (MEMT) is to leverage multiple MT systems to achieve better performance by combining or selecting their outputs. The most popular approach is through a word-level combination framework, such as confusion network (CN) decoding. But it is difficult to incorporate syntactic features in a word-level combination framework because the minimum unit of syntactic analysis is a phrase rather than a word. So we propose to use a phrase or a sentence as the combination unit and thus present a phrase-level combination framework based on paraphrasing processes, and a sentence-level combination technique using consensus and structural language model. Our extensive experiments show significant gains in BLEU over the best single translation engine and the baseline combination system.
BIO
Wei-Yun Ma is a Ph.D. candidate in Computer Science at Columbia University with Professor Kathy McKeown. He received his B.S. in Computer Science from Yuan Ze University and a M.S. in Computer Science at National Chiao Tung University. His primary research interests lie at machine translation, semantic and knowledge representation, question answering and machine learning. He published on top conferences including ACL, NAACL, COLING and AMTA. Before starting graduate study in U.S., he had worked at Institute of Information Science, Academia Sinica, participating in several Chinese NLP projects for many years, including the system development of Chinese word identification, Chinese word sketch engine and the mechanism design for corpus construction.