Raphaël Merx

@ivrik Kat, I'm infinitely sad reading this. May you, Andreas, and your son, find the strength to go through this, and may Lea continue her loving life in a better world.

Meet our baby Lea. She was born with full #Trisomy18 (#Edwards syndrome), a chromosomal abnormality that occurs in 1 in 5,000 pregnancies. 90% of fetuses with this syndrome don't survive to birth, and most of the remaining ones die in the first hours/days/weeks after. Her story🧵

Our #WMT2025 paper got accepted 🙌 We release a dataset for health low-res MT... and find that Gemini 2.5 beats NLLB 54B, if given full document context. Seeya in Suzhou!

Did you know? ❌77% of language models on @huggingface are not tagged for any language 📈For 95% of languages, most models are multilingual 🚨88% of models with tags are trained on English In a new blog post, @tylerachang and I dig into these trends and why they matter! 👇

@davlanade @SenyuLiReal @alabi_jesujoba Really cool work, the AfriDoc-MT paper in particular is a trove of interesting findings on doc Vs sent level translation in low-res MT

5 papers accepted at #EMNLP2025: pp below - SSA-COMET arxiv.org/abs/2506.04557 @SenyuLiReal - AfriDOC-MT arxiv.org/abs/2501.06374 & African NLP survey arxiv.org/abs/2505.21315 @alabi_jesujoba - TransWebLLM arxiv.org/abs/2502.13252 - AfroXLMR-Social arxiv.org/abs/2503.18247

paper: aclanthology.org/2025.acl-demo.… demo: youtu.be/fQFwOxzR4MI Grateful for all the help and advice from my supervisors Hanna Suominen, Nick Thieberger, Trevor Cohn, Katerina Vylomova, and special thanks to Maluk Timor & Lois Hong for the great joint work!

in Vienna for ACL, presenting Tulun, a system for low-resource in-domain translation, using LLMs Working w 2 real use cases: medical translation into Tetun 🇹🇱 & disaster relief speech translation in Bislama 🇻🇺

@YasminMoslem @AdaptCentre @tcddublin @Huawei Congrats Yasmin!

🎉🎉🎉 I have joined @AdaptCentre at Trinity College Dublin @tcddublin as a Research Fellow. In collaboration with @Huawei, we will be making foundation models more efficient, i.e. smaller, faster, and smarter. Stay tuned!

@prajdabre1 must be stressful... I'm sure you'll find something good though, smart driven people like you always land on their feet!

7 days left!

@yanmeg_build there's curse in Cursor for a reason

just watched a neural net hallucinate the entire protestant reformation in 30 seconds. martin luther would've loved github copilot

@prajdabre1 @geoffreyhinton @JeffDean @OriolVinyalsML Whoa that's very cool! Didn't think I'd meet Taylor series approximation again in my lifetime

Paper #6: Distilling the Knowledge in a Neural Network This time I thought I'd summarize a classic. Nothing is more classic than @geoffreyhinton, @JeffDean and @OriolVinyalsML's paper on knowledge distillation. This is one of my favorite reads. Link: arxiv.org/abs/1503.02531

@yanmeg_build Legit good, became aware of the Trudeau resignation through this

exhibit A of a website i'm working on get news through prediction markets north korea running a missile test was unexpected (~13%) until today. cut through the noise, get real news, see what's consensus and what's new pure signal

@jlibovicky Enjoyed reading this, thanks for sharing!

Highlights from multilingual #NLProc and machine translation papers I found on arXiv in November are now on my blog: jlibovicky.github.io/2024/12/05/MTM…

Our paper on generating bilingual example sentences with LLMs got best paper award @altanlp ! arxiv.org/abs/2410.03182 We work with French / Indonesian / Tetun, find that annotators don't agree about what's a "good example", but that LLMs can align with a specific annotator.

@KushalTatariya @akulmizev @EstherPloeger @mmbollmann @johannesbjerva @heather_nlp @mdlhx @lagom_nlp Really nice paper, definitely saving this as inspiration for work on data quality! For §5, you maybe you'd have seen more correlation btw quality & performance on NLG tasks, instead of NLU? As NLG would suffer from low-quality training data, more than NLU, I think?

Our work on quality estimation of non-English Wikipedia articles is on arXiv! 🎉 arxiv.org/abs/2411.05527… In collaboration with @akulmizev, Wessel Poelman, @EstherPloeger, @mmbollmann, @johannesbjerva, Jiaming Luo, @heather_nlp, and @mdlhx at @lagom_nlp✨ 1/5

@zouharvi Cool work, appreciate the video, and fire poster! Can I ask what you used to design it?

About to listen to a song? How about a short trailer for The Pitfalls of Using COMET instead? Approximately the same length. ☄️ youtu.be/jDMvueySuPo

@prajdabre1 @NeurIPSConf Enjoy precious moments with family, congrats to all 3 of you!

@sethjsa @davidstap @diwuNLP @c_monz @illc_amsterdam @ltl_uva Super interesting! In a similar vein, we studied MT using LLMs w sentences from a language manual for Mambai, and found that while scores were decent on test sentences sampled from the manual, it collapsed on sentences translated by a native speaker: aclanthology.org/2024.eurali-1.…

Our work “Can LLMs Really Learn to Translate a Low-Resource Language from One Grammar Book?” is now on arXiv! arxiv.org/abs/2409.19151 - in collaboration with @davidstap, @diwuNLP, @c_monz , and Khalil Sima'an from @illc_amsterdam and @ltl_uva 🧵

@prajdabre1 So cool Raj! Your mom can now say she's made two significant contributions to the NLP community, research and a researcher

This paper is now accepted to the workshop on Asian Translation 2024, collocated with #EMNLP2024! My mom's name will forever be a part of the ACL anthology.

Translation is a complex task involving pre-translation research and post-translation stages. Can #LLMs handle this process step-by-step, relying solely on their internal knowledge? ✨We show that decomposing the translation process significantly improves #Gemini translation quality of long-form texts across all #WMT24 languages! 📜arxiv.org/pdf/2409.06790