Multimodal Neural Machine Translation for English–Assamese Pair

Abstract

Neural machine translation is a state-of-the-art approach for the automatic translation between natural languages. The multimodal concept utilizes textual and image features for improvement in low-resource neural machine translation. There is a lack of a standard multimodal corpus for the English–Assamese low-resource pair. We present a multimodal corpus which is suitable for multimodal translation task of English–Assamese pair. The English–Assamese multimodal corpus is used to implement multimodal neural machine translation models for English-to-Assamese translation and vice-versa. The comparative results of automatic evaluation metrics between text-only and multimodal neural machine translation show multimodal neural machine translation outperforms text-only neural machine translation.