HOME / Exhibition Program / Translating with your favorite expressions
Exhibition Program
| 07 |
Translating with your favorite expressionsLexically constrained neural machine translation 
|
|---|
Although the recent neural machine translation achieves excellent performance, controlling its output expressions is still challenging. We propose a lexically constrained neural machine translation, a method whose translations contain user-specified phrases. Our method improves translation performance while saving inference time and was ranked first in the international competition at WAT 2021. When translating documents in domains such as legal, patent, and scientific, the translation of proper nouns and technical terms is strongly required to be the same expressions throughout the document. Our method will contribute to ensuring consistency in translation by user-specifying expressions.
[1] K. Chousa, M. Morishita, “Input augmentation improves constrained beam search for neural machine translation: NTT at WAT 2021,” in Proc. of the 8th Workshop on Asian Translation (WAT2021), pp. 53–61, 2021.
Katsuki Chousa / Linguistic Intelligence Research Group, Innovative Communication Laboratory
Email: cs-openhouse-ml@hco.ntt.co.jp