Machine Translation to Sign Language Using Post-Translation Replacement Without Placeholders

Sign language is typically the first language for those who are born deaf or who lose their hearing in early childhood. To provide important information for these individuals, it is better to use sign language than to transcribe spoken languages. We have been developing a system that translates Japanese into Japanese Sign Language (JSL) and then generates computer graphics (CG) animation of JSL.In this paper, we propose a machine translation method for translating Japanese into JSL. The proposed method is based on an encoder-decoder model that utilizes a pre-trained model as the encoder, and the proper names in the translation result are revised using a dictionary by means of a post-translation replacement method without placeholders. Our experimental results demonstrate that using the pre-trained model as the encoder and performing the post-translation replacement of proper names both contributed to improving the translation quality.