We are pleased to announce that the CEF Telecom Project User Focused Marian has been officially launched with a successful kick-off meeting.
The University of Edinburgh, Unbabel and Tilde will work towards improving the open-source Marian neural automated translation toolkit. Marian is already widely used in production by eTranslation, Microsoft Translator, World Intellectual Property Organization, US Air Force, eBay among many others. This project will improve Marian tool for automated translation by adding new features commonly requested by its users: factors, forced translation, on-the fly domain adaptation from translation memories, and GPU efficiency.
The project team consists of one academic partner, the University of Edinburgh, and two companies that already use Marian in their products, Unbabel and Tilde. The University of Edinburgh has led Marian development from inception, knows the code well, and brings expertise in optimizing GPU code. Unbabel has been involved in Marian development since the first week, bringing experience on automatic post-editing and the needs of localization companies. Tilde brings a strong C++ engineering team and expertise in terminology management, covering both creation through the eTranslation Termbank project and application for clients.
Together, the consortium team will work to improve Marian toolkit. The action result will be made publicly available through European Language Grid infrastructure and project website not only for use by DSI eTranslation – public administration, but also for industry and language technology developers, facilitating uptake and sustainability after the end of the project.
Stay tuned for news! Meanwhile, if you have any questions, don’t hesitate to reach out to us!