Science of Communication and Computation

Exhibition Program 10

Can computer translate considering context?

Context understanding tests for neural machine translation

Abstract

This project researches methods for evaluating a machine translation system, specifically whether the system could correctly understand the context and produce appropriate translation when multiple sentences were input. We collected examples of correct translations in which their translations changed when preceding sentences were different. They were then classified according to their linguistic phenomenon, which is the key to understanding different meanings. Tests of contextual understanding were then performed. We have created the first database that systematically collects examples that are necessary for understanding context when translating from Japanese to English. In a Japanese sentence, the subject or object is often omitted. Different meanings of the same word are represented by different kanji characters. We therefore need a novel approach to Japanese-English translation that is different from previous research on English to French translation.

Reference

  • [1] M. Morishita, J. Suzuki, M. Nagata, “NTT Neural Machine Translation Systems at WAT 2017,” in Proc. The 4th Workshop on Asian Translation (WAT-2017), 2017.
    [2] R, Bawden, R. Wennrich, A. Birch, B. Haddow, “Evaluating Discourse Phenomena in Neural Machine Translation,” in Proc. The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL-2018), 2018.

Poster

Photos

Presenters

Masaaki Nagata
Masaaki Nagata
Innovative Communication Laboratory