There are many different forms of comics around the world, such as manga in Japan, We devised a first comprehensive system for fully automated manga translation. Third, we created a new benchmark to evaluate manga translation. Second, for training the model, we propose the approach to automaticĬorpus construction from pairs of original manga and their translations,īy which a large parallel corpus can be constructed without any manual labeling. Other speech bubbles, gender of speakers, etc.). That cannot be translated without using context information (e.g., texts in It enables us to translate texts in speech bubbles We are the first to incorporate context information obtained from manga images. In this paper, we make the following four contributions thatĮstablish the foundation of manga translation research.įirst, we propose a multimodal context-aware translation framework. In addition, corpus and benchmarks to train and evaluate such models are currently unavailable. However, it is still an open problem how to extract context from images and integrate into MT models. Obtaining context from the image is essential for its translation. Since text and images are mixed up in an unstructured fashion in manga, Manga translation involves two important problems in MT: context-aware and multimodal translation. We tackle the problem of machine translation (MT) of manga, Japanese comics.
