Adding Visual Information to Improve Multimodal Machine Translation for Low-Resource Language

Machine translation makes it easy for people to communicate across languages. Multimodal machine is also one of the important directions research in translation, which uses feature information such as images and audio assist models obtaining higher quality target However, vast majority current work has been conducted on basis commonly used corpora English, French, German, less done low-resource languages, this left languages relatively behind. This paper selects English-Hindi English-Hausa corpus, researched language translation. The different we use image extraction are fusion features with text encoding process using provide additional information, assisting model Compared text-only experimental results show that our method improves 3 BLEU dataset 0.47 dataset. In addition, analyze effect extracted by results. Different pay attention each region image, ResNet able extract more compared VGG model, effective