Braille translation system using neural machine translation technology I - code conversion

Abstract

We have constructed a translation system to use easily even for beginners. In order to translate Japanese into Braille, we programmed in Python and C language using MeCab, a morphological analysis engine. However, there are many exceptions in Braille grammar, and the programming has become complicated. In recent years, researches on the machine translation using a neural network are thriving, so we carried out braille translation using the technology of neural machine translation (NMT) this time. Tensorflow was used as a library of neural network. We also incorporated a transform which was a translation component in the open source library Tensor2Tensor. We used 100,000 words of Japanese as teacher data. By using NMT, the morphological engine and the complicated programming become unnecessary. Also, MNT can translate without worrying about the complex Braille grammar. Then, just by entering Japanese sentence, the Braille codes are output. There are many advantages. We describe the braille translation using NMT.