In this paper, we propose a reinforcement learning approach to unseen language translation given<br>rich linguistic context, we argue that LLMs can acquire the meta-skill of utilizing context knowledge rather than memorizing specific languages thus have generalization ability to unseen languages.</p>\n","updatedAt":"2026-06-05T03:38:50.712Z","author":{"_id":"63898876f7d3b0df09221ed0","avatarUrl":"/avatars/f7a9fe8a7be30bd07a10df2b286a0711.svg","fullname":"Hanxu Hu","name":"HanxuHU","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":8,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8370423316955566},"editors":["HanxuHU"],"editorAvatarUrls":["/avatars/f7a9fe8a7be30bd07a10df2b286a0711.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.06428","authors":[{"_id":"6a2242983490a593e87b1502","name":"Hanxu Hu","hidden":false},{"_id":"6a2242983490a593e87b1503","name":"Zdeněk Šnajdr","hidden":false},{"_id":"6a2242983490a593e87b1504","name":"Pinzhen Chen","hidden":false},{"_id":"6a2242983490a593e87b1505","name":"Jannis Vamvas","hidden":false},{"_id":"6a2242983490a593e87b1506","name":"Rico Sennrich","hidden":false}],"publishedAt":"2026-06-04T00:00:00.000Z","submittedOnDailyAt":"2026-06-05T00:00:00.000Z","title":"Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation","submittedOnDailyBy":{"_id":"63898876f7d3b0df09221ed0","avatarUrl":"/avatars/f7a9fe8a7be30bd07a10df2b286a0711.svg","isPro":true,"fullname":"Hanxu Hu","user":"HanxuHU","type":"user","name":"HanxuHU"},"summary":"Prior work has shown that large language models (LLMs) can translate unseen or low-resource languages by undergoing continued training or even by encoding a grammar book in their context. However, both methods typically overfit specific languages, with limited zero-shot transfer at test time. To translate extremely low-resource languages at scale, we argue that LLMs must acquire the meta-skill of utilizing in-context linguistic knowledge rather than memorizing specific languages. In this paper, we propose a reinforcement learning (RL) approach to unseen language translation given rich linguistic context, using a surface-level translation metric (chrF) as the reward. Empirically, despite the lightweight reward, our RL-trained models effectively extract and apply relevant linguistic information from the provided context, leading to better translations on completely unseen languages than in-context learning or supervised fine-tuning. Our analyses suggest that outcome-based RL can extend beyond conventional reasoning tasks like math and coding to serve as a recipe for language learning from context.","upvotes":20,"discussionId":"6a2242993490a593e87b1507","githubRepo":"https://github.com/hanxuhu/rl-new-language","githubRepoAddedBy":"user","ai_summary":"Reinforcement learning approach enables large language models to translate unseen languages by leveraging in-context linguistic knowledge rather than memorizing specific languages.","ai_keywords":["large language models","reinforcement learning","in-context learning","supervised fine-tuning","chrF","linguistic context","zero-shot transfer","meta-skill"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":2,"organization":{"_id":"64007d5df4ff62c2616d011e","name":"ZurichNLP","fullname":"University of Zurich, Department of Computational Linguistics","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1677753608204-622a41f973f278bb2fb1c6a4.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63bc77661374e3ef9135735f","avatarUrl":"/avatars/94b04545ed9d30bfe58691672a0b5618.svg","isPro":false,"fullname":"Qingcheng Zeng","user":"qcz","type":"user"},{"_id":"63898876f7d3b0df09221ed0","avatarUrl":"/avatars/f7a9fe8a7be30bd07a10df2b286a0711.svg","isPro":true,"fullname":"Hanxu Hu","user":"HanxuHU","type":"user"},{"_id":"6560763e152b659e623865ae","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6560763e152b659e623865ae/cTT2jGnPU_8XMrUTvqZ2h.jpeg","isPro":false,"fullname":"Xiao Liang","user":"MasterVito","type":"user"},{"_id":"65eaf755ab0a6a90da55ab58","avatarUrl":"/avatars/a46890a9d067a913513edf3759f12c85.svg","isPro":false,"fullname":"Cunxiang Wang","user":"wangcunxiang","type":"user"},{"_id":"61669c456916c52acd5a1aa3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61669c456916c52acd5a1aa3/HnZTwRaXgTeTG3ljO3ITb.jpeg","isPro":false,"fullname":"jianbo dai","user":"jbd","type":"user"},{"_id":"6589aa6cd2ea3f329459f27b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6589aa6cd2ea3f329459f27b/koQzvDtYiR_RJ_801zD-H.jpeg","isPro":false,"fullname":"Chenmien Tan","user":"Chenmien","type":"user"},{"_id":"636681feaa6a4af6073ba73e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/636681feaa6a4af6073ba73e/u_0moYNu6as9Sszp-ej95.png","isPro":true,"fullname":"Simon Yu","user":"simonycl","type":"user"},{"_id":"67e95c8d2b124840d0cb8d7f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/UCt1iaTveaIXA-NBEqX3A.png","isPro":false,"fullname":"shawnxzhu","user":"shawnxzhu","type":"user"},{"_id":"63b6af3accebeadccc868efd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b6af3accebeadccc868efd/cFTHKggMpsoaPe_46gcy9.webp","isPro":false,"fullname":"Zhijiang","user":"Zeee","type":"user"},{"_id":"67f8ccce9301e8cd1592b71f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/pmQUnMY4GqTYmm0K-_7BA.png","isPro":false,"fullname":"WangZilin","user":"terr1ble","type":"user"},{"_id":"68e516daf823014ee57ba199","avatarUrl":"/avatars/bf4d3efb90e199250b18b9eac6657188.svg","isPro":false,"fullname":"Yuxuan Wang","user":"Yuxuan566","type":"user"},{"_id":"64fed23f0871bc5930598ab5","avatarUrl":"/avatars/080a4ef3e4634cd978528dfa899a4eb0.svg","isPro":false,"fullname":"ZhiWei LI","user":"Aragonaa","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"64007d5df4ff62c2616d011e","name":"ZurichNLP","fullname":"University of Zurich, Department of Computational Linguistics","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1677753608204-622a41f973f278bb2fb1c6a4.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.06428.md"}">
Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation
Abstract
Reinforcement learning approach enables large language models to translate unseen languages by leveraging in-context linguistic knowledge rather than memorizing specific languages.
Prior work has shown that large language models (LLMs) can translate unseen or low-resource languages by undergoing continued training or even by encoding a grammar book in their context. However, both methods typically overfit specific languages, with limited zero-shot transfer at test time. To translate extremely low-resource languages at scale, we argue that LLMs must acquire the meta-skill of utilizing in-context linguistic knowledge rather than memorizing specific languages. In this paper, we propose a reinforcement learning (RL) approach to unseen language translation given rich linguistic context, using a surface-level translation metric (chrF) as the reward. Empirically, despite the lightweight reward, our RL-trained models effectively extract and apply relevant linguistic information from the provided context, leading to better translations on completely unseen languages than in-context learning or supervised fine-tuning. Our analyses suggest that outcome-based RL can extend beyond conventional reasoning tasks like math and coding to serve as a recipe for language learning from context.
Community
In this paper, we propose a reinforcement learning approach to unseen language translation given
rich linguistic context, we argue that LLMs can acquire the meta-skill of utilizing context knowledge rather than memorizing specific languages thus have generalization ability to unseen languages.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2606.06428 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.06428 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.