BakaTranslator is a multimodal deep learning based JA -> EN translation tool.
+-------------------------------------------------------------------------------------------------------+
| Warning: Do not use this model for ANY sources you do not have explicit permission to use. |
| The author of this repository is not responsible for any misuse of the model. |
| This project is not available for commercial use. |
+-------------------------------------------------------------------------------------------------------+
Translation practices generally fall into two broad philosophies: literal (word-for-word) translation and contextual translation that accounts for subtext and nuance - also known as localization. While literal translation is often more straightforward to execute, this project is grounded in the latter approach, prioritizing the accurate conveyance of underlying meaning, cultural context, and implied intent.
This approach is particularly important when translating from Japanese to English. The Japanese language often relies on implied meaning, with key elements such as subjects, pronouns, or emotions left unstated but understood within context. A direct translation may miss these cues, resulting in language that feels flat, awkward, or incomplete in English. By focusing on subtext, this project seeks to produce translations that are not only accurate but also natural and faithful to the original's deeper meaning.
Some Japanese phrases are incomprehensible/Untranslatable to English - how to tackle?
Data is unstructured, and more importantly, split
Hinami, Ryota, et al. Towards Fully Automated Manga Translation. arXiv, 2020, https://arxiv.org/abs/2012.14271.
Matsui, Yusuke, et al. "Sketch-Based Manga Retrieval Using Manga109 Dataset." Multimedia Tools and Applications, vol. 76, no. 20, 2017, pp. 21811–21838. https://doi.org/10.1007/s11042-016-4020-z.
Aizawa, Kiyoharu, et al. "Building a Manga Dataset 'Manga109' with Annotations for Multimedia Applications." IEEE MultiMedia, vol. 27, no. 2, 2020, pp. 8–18. https://doi.org/10.1109/mmul.2020.2987895.
This project is not possible without The Aizawa, Yamakata, Matsui Lab (HAL) Lab from the University of Tokyo and The MANTRA team.
All images are owned by the respective copyright holders. All rights reserved