Login
Computer Science @ Rochester
Tuesday, July 27, 2010
2:00 PM
Computer Studies Bldg. Room 632
Ph.D. Thesis Proposal
Tagyoung Chung
University of Rochester
Translating Morphologically Rich Languages
Morphologically rich languages present multiple challenges to machine translation compared to more frequently studied languages such as Chinese and English because these languages possess many characteristics that are overlooked in current machine translation research. Morphologically rich languages are often written in a way that requires tokenization of text before it can be analyzed. Null anaphora is more prevalent in these languages, which creates problems when translating into languages that have no such phenomenon. Although generating the right morphology is essential for conveying the correct meaning of sentences, it is often neglected.

In this thesis proposal, we examine current research on handling morphologically rich languages in statistical machine translation and show our work. We start by introducing recent work on unsupervised tokenization and present our work on unsupervised tokenization specifically designed for machine translation. We also present our experiments with null elements and show analysis of how they affect parsing and translation. We also discuss how our work on integrating null anaphora can be improved using coreference resolution. We finally propose a semantically-inspired method of generating morphology when the target language is rich in morphology.