Addressing two problems in neural machine translation decoding

Release Time:2019-04-19Number of visits:150

Speaker:    Prof.Yue Zhang

Time:       9:30-10:30, May. 6

Location:    SIST 1A 200

Host:         Prof. Kewei Tu

Abstract:

I will talk about two projects that fix problems in neural machine translation decoding, which are done collaboratively with Kai Song. The first addresses sparsity in the vocabulary. In particular, for English to Russian translation, the output words are highly morphological. We tried using a separate module for generating morphology and observed much better BLEU. The second addresses issues for forced translation. In practice a user may have a lexicon from which a set of fixed terms have their preferred translation. We study data augmentation for enforcing lexicon translation, finding that it gives better effectiveness than two standard methods.

Bio:

Yue Zhang is currently an associate professor at Westlake University. His research interests include natural language processing and computational finance. He has been working on statistical parsing, text synthesis, natural language synthesis, machine translation, information extraction, sentiment analysis and stock market analysis intensively. He won the best paper awards of IALP 2017 and COLING 2018. Yue Zhang serves the editorial board for Transaction of Association of Computational Linguistics, ACM Transactions on Asian and Low Resource Language Information Processing (associate editor) and IEEE Transactions on Big Data (associate editor), and as area chairs of COLING 2014/18, NAACL 2015/19, EMNLP 2015/17/19, ACL 2017/18/19.

SIST-Seminar 18146