User-Focused Marian Name: User-Focused Marian Coordinator: Kenneth Heafield, UEDIN Consortium: University of Edinburgh, Unbabel Lda, Tilde SIA Project Runtime: 1 March 2020 – 28 February 2022 Funded by: European Commission https://marian-project.eu/ The project aims at improving the open-source neural machine translation toolkit Marian with commonly requested features: factors, forced translation, on-the fly domain adaptation from translation memories, and GPU efficiency. In particular the project aims to provide user access to text flagging by adding support for factors to support forced translation. It will also support the automated adaptation facility by adding for on-the-fly domain adaptation using large translation memories with a well-defined API. The speed improvement through increased efficiency on GPUs is planned. Finally, Marian’s documentation will be improved for all levels of users.