{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,2]],"date-time":"2026-09-02T20:21:58Z","timestamp":1788380518977,"version":"build-2803163510"},"reference-count":52,"publisher":"MIT Press","license":[{"start":{"date-parts":[[2026,9,2]],"date-time":"2026-09-02T00:00:00Z","timestamp":1788307200000},"content-version":"vor","delay-in-days":244,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["direct.mit.edu"],"crossmark-restriction":true},"short-container-title":[],"published-print":{"date-parts":[[2026,8,28]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Large reasoning models (LRMs) have shown exceptional performance in complex tasks such as mathematics and coding. In the field of machine translation (MT), reinforcement learning (RL) has been utilized to enhance the quality of translations. However, traditional RL approaches rely heavily on the base model\u2019s inherent translation capabilities, which may falter when dealing with terminology translations and domain-specific expressions without sufficient guidance. In this paper, we introduce FeedTrans (Feedback-driven Translation), which employs a FeedRollout mechanism to incorporate feedback as guidance, enabling the production of high-quality translations and expanding the search space for improved translation outcomes. Extensive experiments across six translation tasks validate the effectiveness of our approach. To the best of our knowledge, this is the first attempt at utilizing feedback in MT-oriented RL. https:\/\/github.com\/Acerkoo\/FeedTrans<\/jats:p>","DOI":"10.1162\/tacl.a.784","type":"journal-article","created":{"date-parts":[[2026,9,2]],"date-time":"2026-09-02T20:01:01Z","timestamp":1788379261000},"page":"1899-1916","update-policy":"https:\/\/doi.org\/10.1162\/mitpressjournals.corrections.policy","source":"Crossref","is-referenced-by-count":0,"title":["FeedTrans: Enhance Machine Translation via Feedback"],"prefix":"10.1162","volume":"14","author":[{"given":"Zengkui","family":"Sun","sequence":"first","affiliation":[{"name":"Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing Jiaotong University, Beijing, China, zengksun@bjtu.edu.cn"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiali","family":"Zeng","sequence":"additional","affiliation":[{"name":"WeChat AI, Tencent Inc, China, lemonzeng@tencent.com"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiaan","family":"Wang","sequence":"additional","affiliation":[{"name":"WeChat AI, Tencent Inc, China, torchwang@tencent.com"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fandong","family":"Meng","sequence":"additional","affiliation":[{"name":"WeChat AI, Tencent Inc, China, fandongmeng@tencent.com"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xinyan","family":"Guan","sequence":"additional","affiliation":[{"name":"WeChat AI, Tencent Inc, China, guanxy0406@gmail.com"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yufeng","family":"Chen","sequence":"additional","affiliation":[{"name":"Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing Jiaotong University, Beijing, China, chenyf@bjtu.edu.cn"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jinan","family":"Xu","sequence":"additional","affiliation":[{"name":"Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing Jiaotong University, Beijing, China, jaxu@bjtu.edu.cn"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wenjuan","family":"Han","sequence":"additional","affiliation":[{"name":"Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing Jiaotong University, Beijing, China, wjhan@bjtu.edu.cn"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jie","family":"Zhou","sequence":"additional","affiliation":[{"name":"WeChat AI, Tencent Inc, China, withtomzhou@tencent.com"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"281","published-online":{"date-parts":[[2026,8,28]]},"reference":[{"key":"2026090216005536300_bib1","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2020.nlpcovid19-2.5","article-title":"Tico-19: the translation initiative for covid-19","volume-title":"arXiv preprint arXiv:2007.01788","author":"Anastasopoulos","year":"2020"},{"key":"2026090216005536300_bib2","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2024.acl-short.64","article-title":"Dual-reflect: Enhancing large language models for reflective translation through dual learning feedback mechanisms","volume-title":"arXiv preprint arXiv:2406.07232","author":"Chen","year":"2024"},{"key":"2026090216005536300_bib3","article-title":"Evaluating o1-like llms: Unlocking reasoning for translation through comprehensive analysis","volume-title":"arXiv preprint arXiv:2502.11544","author":"Chen","year":"2025"},{"key":"2026090216005536300_bib4","article-title":"Iterative translation refinement with large language models","volume-title":"arXiv preprint arXiv:2306.03856","author":"Chen","year":"2023"},{"key":"2026090216005536300_bib5","article-title":"Towards reasoning era: A survey of long chain-of-thought for reasoning large language models","volume-title":"arXiv preprint arXiv:2503.09567","author":"Chen","year":"2025"},{"key":"2026090216005536300_bib6","article-title":"No language left behind: Scaling human-centered machine translation","volume-title":"arXiv preprint arXiv:2207.04672","author":"Costa-Juss\u00e0","year":"2022"},{"key":"2026090216005536300_bib7","article-title":"Retool: Reinforcement learning for strategic tool use in llms","volume-title":"arXiv preprint arXiv:2504.11536","author":"Feng","year":"2025"},{"key":"2026090216005536300_bib8","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2025.findings-emnlp.1015","article-title":"Mt-r1-zero: Advancing llm-based machine translation via r1-zero-like reinforcement learning","volume-title":"arXiv preprint arXiv:2504.10160","author":"Feng","year":"2025"},{"key":"2026090216005536300_bib9","doi-asserted-by":"crossref","first-page":"522","DOI":"10.1162\/tacl_a_00474","article-title":"The flores-101 evaluation benchmark for low-resource and multilingual machine translation","volume":"10","author":"Goyal","year":"2022","journal-title":"Transactions of the Association for Computational Linguistics"},{"key":"2026090216005536300_bib10","article-title":"The llama 3 herd of models","volume-title":"arXiv preprint arXiv:2407.21783","author":"Grattafiori","year":"2024"},{"key":"2026090216005536300_bib11","article-title":"Deeprag: Thinking to retrieval step by step for large language models","volume-title":"arXiv preprint arXiv:2502.01142","author":"Guan","year":"2025"},{"key":"2026090216005536300_bib12","article-title":"Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning","volume-title":"arXiv preprint arXiv:2501.12948","author":"Guo","year":"2025"},{"key":"2026090216005536300_bib13","volume-title":"Redefining machine translation on social network services with large language models","author":"Guo","year":"2025"},{"key":"2026090216005536300_bib14","article-title":"R1-t1: Fully incentivizing translation capability in llms via reasoning learning","volume-title":"arXiv preprint arXiv:2502.19735","author":"He","year":"2025"},{"key":"2026090216005536300_bib15","volume-title":"Open-reasoner-zero: An open source approach to scaling up reinforcement learning on the base model","author":"Hu","year":"2025"},{"key":"2026090216005536300_bib16","article-title":"Think-j: Learning to think for generative llm-as-a-judge","volume-title":"arXiv preprint arXiv:2505.14268","author":"Huang","year":"2025"},{"key":"2026090216005536300_bib17","article-title":"Openai o1 system card","volume-title":"arXiv preprint arXiv:2412.16720","author":"Jaech","year":"2024"},{"key":"2026090216005536300_bib18","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/P17-2074","article-title":"Best-worst scaling more reliable than rating scales: A case study on sentiment intensity annotation","volume-title":"arXiv preprint arXiv:1712.01765","author":"Kiritchenko","year":"2017"},{"key":"2026090216005536300_bib19","doi-asserted-by":"crossref","DOI":"10.1145\/3600006.3613165","article-title":"Efficient memory management for large language model serving with pagedattention","volume-title":"Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles","author":"Kwon","year":"2023"},{"key":"2026090216005536300_bib20","doi-asserted-by":"crossref","first-page":"6445","DOI":"10.18653\/v1\/2024.naacl-long.358","article-title":"Mt-patcher: Selective and extendable knowledge distillation from large language models for machine translation","volume-title":"Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","author":"Li","year":"2024"},{"key":"2026090216005536300_bib21","article-title":"Torl: Scaling tool-integrated rl","volume-title":"arXiv preprint arXiv:2503.23383","author":"Li","year":"2025"},{"key":"2026090216005536300_bib22","article-title":"From system 1 to system 2: A survey of reasoning large language models","volume-title":"arXiv preprint arXiv:2502.17419","author":"Li","year":"2025"},{"key":"2026090216005536300_bib23","article-title":"Slangdit: Benchmarking llms in interpretative slang translation","volume-title":"arXiv preprint arXiv:2505.14181","author":"Liang","year":"2025"},{"key":"2026090216005536300_bib24","article-title":"Deepseek-v3 technical report","volume-title":"arXiv preprint arXiv:2412.19437","author":"Liu","year":"2024"},{"key":"2026090216005536300_bib25","article-title":"New trends for modern machine translation with large reasoning models","volume-title":"arXiv preprint arXiv:2503.10351","author":"Liu","year":"2025"},{"key":"2026090216005536300_bib26","volume-title":"There may not be aha moment in r1-zero-like training \u2014 a pilot study","author":"Liu","year":"2025"},{"key":"2026090216005536300_bib27","volume-title":"Deepscaler: Surpassing o1-preview with a 1.5b model by scaling rl","author":"Luo","year":"2025"},{"key":"2026090216005536300_bib28","doi-asserted-by":"crossref","DOI":"10.52202\/085713-3524","article-title":"Toolrl: Reward is all tool learning needs","volume-title":"arXiv preprint arXiv:2504.13958","author":"Qian","year":"2025"},{"key":"2026090216005536300_bib29","doi-asserted-by":"crossref","first-page":"3505","DOI":"10.1145\/3394486.3406703","article-title":"Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters","volume-title":"Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining","author":"Rasley","year":"2020"},{"key":"2026090216005536300_bib30","article-title":"Leveraging gpt-4 for automatic translation postediting","volume-title":"arXiv preprint arXiv:2305.14878","author":"Raunak","year":"2023"},{"key":"2026090216005536300_bib31","doi-asserted-by":"crossref","first-page":"578","DOI":"10.18653\/v1\/2022.wmt-1.52","article-title":"Comet-22: Unbabel-ist 2022 submission for the metrics shared task","volume-title":"Proceedings of the Seventh Conference on Machine Translation (WMT)","author":"Rei","year":"2022"},{"key":"2026090216005536300_bib32","doi-asserted-by":"crossref","first-page":"2685","DOI":"10.18653\/v1\/2020.emnlp-main.213","article-title":"COMET: A neural framework for MT evaluation","volume-title":"Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)","author":"Rei","year":"2020"},{"key":"2026090216005536300_bib33","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.wmt-1.60","article-title":"Cometkiwi: Ist-unbabel 2022 submission for the quality estimation shared task","volume-title":"arXiv preprint arXiv:2209.06243","author":"Rei","year":"2022"},{"key":"2026090216005536300_bib34","article-title":"Deepseekmath: Pushing the limits of mathematical reasoning in open language models","volume-title":"arXiv preprint arXiv:2402.03300","author":"Shao","year":"2024"},{"key":"2026090216005536300_bib35","article-title":"Hybridflow: A flexible and efficient rlhf framework","volume-title":"arXiv preprint arXiv: 2409.19256","author":"Sheng","year":"2024"},{"key":"2026090216005536300_bib36","article-title":"Crossing the reward bridge: Expanding rl with verifiable rewards across diverse domains","volume-title":"arXiv preprint arXiv:2503.23829","author":"Su","year":"2025"},{"key":"2026090216005536300_bib37","article-title":"Kimi k1. 5: Scaling reinforcement learning with llms","volume-title":"arXiv preprint arXiv:2501.12599","author":"Team","year":"2025"},{"key":"2026090216005536300_bib38","article-title":"Otc: Optimal tool calls via reinforcement learning","volume-title":"arXiv preprint arXiv:2504.14870","author":"Wang","year":"2025"},{"key":"2026090216005536300_bib39","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2023.newsum-1.1","article-title":"Is chatgpt a good nlg evaluator? a preliminary study","volume-title":"arXiv preprint arXiv:2303.04048","author":"Wang","year":"2023"},{"key":"2026090216005536300_bib40","article-title":"Drt-o1: Optimized deep reasoning translation via long chain-of-thought","volume-title":"arXiv e-prints","author":"Wang","year":"2024"},{"key":"2026090216005536300_bib41","article-title":"Deep reasoning translation via reinforcement learning","volume-title":"arXiv preprint arXiv:2504.10187","author":"Wang","year":"2025"},{"key":"2026090216005536300_bib42","article-title":"Extrans: Multilingual deep reasoning translation via exemplar-enhanced reinforcement learning","volume-title":"arXiv preprint arXiv:2505.12996","author":"Wang","year":"2025"},{"key":"2026090216005536300_bib43","article-title":"Delta: An online document-level translation agent based on multi-level memory","volume-title":"arXiv preprint arXiv:2410.08143","author":"Wang","year":"2024"},{"key":"2026090216005536300_bib44","doi-asserted-by":"crossref","DOI":"10.52202\/085713-3904","article-title":"Learning to reason under off-policy guidance","volume-title":"arXiv preprint arXiv:2504.14945","author":"Yan","year":"2025"},{"key":"2026090216005536300_bib45","volume-title":"arXiv preprint arXiv:2412.15115","author":"Yang","year":"2024"},{"key":"2026090216005536300_bib46","article-title":"Ssr-zero: Simple self-rewarding reinforcement learning for machine translation","volume-title":"arXiv preprint arXiv:2505.16637","author":"Yang","year":"2025"},{"key":"2026090216005536300_bib47","article-title":"Benchmarking machine translation with cultural awareness","volume-title":"arXiv preprint arXiv:2305.14328","author":"Yao","year":"2023"},{"key":"2026090216005536300_bib48","doi-asserted-by":"crossref","DOI":"10.52202\/085713-3775","article-title":"Dapo: An open-source llm reinforcement learning system at scale","volume-title":"arXiv preprint arXiv:2503.14476","author":"Yu","year":"2025"},{"key":"2026090216005536300_bib49","article-title":"Does reinforcement learning really incentivize reasoning capacity in llms beyond the base model?","volume-title":"arXiv preprint arXiv:2504.13837","author":"Yue","year":"2025"},{"key":"2026090216005536300_bib50","article-title":"o1-coder: an o1 replication for coding","volume-title":"arXiv preprint arXiv:2412.00154","author":"Zhang","year":"2024"},{"key":"2026090216005536300_bib51","article-title":"Echo chamber: Rl post-training amplifies behaviors learned in pretraining","volume-title":"arXiv preprint arXiv:2504.07912","author":"Zhao","year":"2025"},{"key":"2026090216005536300_bib52","article-title":"Marco-o1: Towards open reasoning models for open-ended solutions","volume-title":"arXiv preprint arXiv:2411.14405","author":"Zhao","year":"2024"}],"container-title":["Transactions of the Association for Computational Linguistics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/direct.mit.edu\/tacl\/article-pdf\/doi\/10.1162\/TACL.a.784\/2624567\/tacl.a.784.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/direct.mit.edu\/tacl\/article-pdf\/doi\/10.1162\/TACL.a.784\/2624567\/tacl.a.784.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,2]],"date-time":"2026-09-02T20:01:15Z","timestamp":1788379275000},"score":1,"resource":{"primary":{"URL":"https:\/\/direct.mit.edu\/tacl\/article\/doi\/10.1162\/TACL.a.784\/138655\/FeedTrans-Enhance-Machine-Translation-via-Feedback"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":52,"URL":"https:\/\/doi.org\/10.1162\/tacl.a.784","relation":{},"ISSN":["2307-387X"],"issn-type":[{"value":"2307-387X","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026]]},"published":{"date-parts":[[2026]]}}}