MetaDocEval: A Contrastive Framework for Evaluating Machine Translation Metrics at the Document-Level Nicolas Dahan, Rachel Bawden and François Yvon 📅 18th June | 09:05-09:30 | oral

🔹 Main Conference The MaTOS Pipeline for the Translation of Scientific Abstracts on the HAL Platform Panagiotis Tsolakis, Ziqian Peng, Laurent Romary, François Yvon and Rachel Bawden 📅 17th June | 15:30-15:55 | oral

Accepted to a Workshop (1/2): "Self-Retrieval from Distant Contexts for Document-Level Machine Translation", accepted to the Conference on Machine Translation (WMT25), from @ziqianpeng.bsky.social, @rachelbawden.bsky.social, @yvofr.bsky.social

We will also announce a new experiment on post-editing abstracts in the #NLP domain. If you can post-edit into French, please register and donate your post-edit to science ! Details on our website: urls.fr/7zlHtZ.

Very happy to be in Geneva for the #MTSummit2025. Tomorrow we present our recent progresses on the translation of scholarly documents. Check our resources and publications on the project website (anr-matos.github.io).

1: Towards the Machine Translation of Scientific Neologisms with @yvofr.bsky.social : ever struggled to translate a new term such as pretraining or Reinforcement Learning from Human Feedback? We aim to leverage the definitions of terms to translate them more accurately

2: Unlike “Likely”, “Unlike” is Unlikely: always with @yvofr.bsky.social Because BPE makes a difference between tokens at the beginning and the end of words, LLMs are unable to generate prefixations