Elamur Research⌕
☰
home / papers / 2026 / LaCoMSA: Language-Consistency Multilingual Self-Alignment with Latent Representation Rewarding

LaCoMSA: Language-Consistency Multilingual Self-Alignment with Latent Representation Rewarding

Paper · 2026 · Khanh-Tung Tran, Barry O'Sullivan and Hoang D. Nguyen

Summary

Multilingual LLMs tend to answer worse in lower-resource languages than in English. They also often switch into English even when prompted in another language. Earlier preference-optimization fixes such as MAPO, LIDR and ICR depend on outside supervision, such as translation systems or English-trained reward models.

The authors instead build rewards from the model's own hidden states. In MSA, the model answers the same prompt in English and in a target language. The non-English answer is scored by its average cosine similarity to the English answers at a middle "semantic hub" layer, layer 15 in the Llama 3 8B experiments. LaCoMSA adds a last-layer factor that compares the output against English and target-language centroids, which lowers the reward for answers that drift into English. The resulting preference pairs train the model with DPO and Iterative DPO.

On X-AlpacaEval, LaCoMSA lifts the length-controlled win rate of Llama-3-8B-SFT-DPO by up to 6.8 points (55.0% relative), with consistent gains on other benchmarks and models. Since it needs no translators, annotations or external embedding models, the method suits languages that lack such tooling.

Key points

  • Silhouette analysis used 100 UltraFeedback questions with 10 sampled answers per language. Answers group by question most strongly in the middle layers and much less at the final layer. The semantic hub therefore appears even with self-generated parallel answers rather than parallel corpora.
  • Experiments covered Spanish, Russian, German and French. The default training set was 3,000 prompts per language. With only 1,000 prompts, LaCoMSA still gained 4.58 points (37.33% relative) on X-AlpacaEval.
  • Language fidelity was checked with the NLLB language identifier. On X-AlpacaEval, LaCoMSA raised the average correct-language rate from 80.62% to 90.78%. Russian rose from 70.93% to 83.98%. MSA alone collapsed to roughly 20% average by iteration 2.
  • Ablation: MSA without the consistency factor helps on MT-Bench but hurts on X-AlpacaEval because of off-target English replies. LaCoMSA improves results across all benchmarks.
  • General knowledge was largely preserved. Multilingual ARC, HellaSwag and TruthfulQA changed only marginally, and Multilingual MMLU rose from 0.5349 to about 0.552–0.553. Middle-layer clustering also strengthened, for example en+ru silhouette rose from 0.3208 to 0.3639. Code, weights and data are available at github.com/ReML-AI/LaCoMSA.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🔬Insight SFI Research Centre for Data AnalyticsDublin · Ireland9 works · 543 citations🎓University College CorkCork · Ireland5 works · 24 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Khanh-Tung Tran, Barry O'Sullivan and Hoang D. Nguyen. 2026. LaCoMSA: Language-Consistency Multilingual Self-Alignment with Latent Representation Rewarding. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-long.224

@inproceedings{tran2026lacomsa,
  title = {{LaCoMSA: Language-Consistency Multilingual Self-Alignment with Latent Representation Rewarding}},
  author = {Khanh-Tung Tran, Barry O'Sullivan and Hoang D. Nguyen},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-long.224},
  url = {https://doi.org/10.18653/v1/2026.eacl-long.224}
}

References

44 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. AI@Meta. 2024. Llama 3 model card. find
  2. Firoj Alam, Shammur Absar Chowdhury, Sabri Boughorbel, and Maram Hasanain. 2024. LLMs for low resource languages in multilingual, multimodal and dialectal settings. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: Tutorial Abstracts, pag… find
  3. Niyati Bafna, Tianjian Li, Kenton Murray, David R. Mortensen, David Yarowsky, Hale Sirin, and Daniel Khashabi. 2025. The translation barrier hypothesis: Multilingual generation with large language models suffers from implicit translation failure. Preprint, arXiv:2506.22724.
  4. Lichang Chen, Shiyang Li, Jun Yan, Hai Wang, Kalpa Gunaratna, Vikas Yadav, Zheng Tang, Vijay Srinivasan, Tianyi Zhou, Heng Huang, and Hongxia Jin. 2024. Alpagasus: Training a better alpaca with fewer data. In The Twelfth International Conference on Learning Representations.
  5. Zewen Chi, Li Dong, Furu Wei, Nan Yang, Saksham Singhal, Wenhui Wang, Xia Song, Xian-Ling Mao, Heyan Huang, and Ming Zhou. 2021. InfoXLM: An information-theoretic framework for cross-lingual language model pre-training. In Proceedings of the 2021 Conference of the North American Chapter of the Assoc…
  6. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. Preprint, arXiv:1803.05457.
  7. Ganqu Cui, Lifan Yuan, Ning Ding, Guanming Yao, Wei Zhu, Yuan Ni, Guotong Xie, Zhiyuan Liu, and Maosong Sun. 2024. Ultrafeedback: Boosting language models with high-quality feedback. find
  8. Ning Ding, Yulin Chen, Bokai Xu, Yujia Qin, Shengding Hu, Zhiyuan Liu, Maosong Sun, and Bowen Zhou. 2023. Enhancing chat language models by scaling high-quality instructional conversations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 3029–3051, Si…
  9. Fangxiaoyu Feng, Yinfei Yang, Daniel Cer, Naveen Arivazhagan, and Wei Wang. 2022. Language-agnostic BERT sentence embedding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 878–891, Dublin, Ireland. Association for Computation…
  10. Jiatao Gu, Yong Wang, Kyunghyun Cho, and Victor O.K. Li. 2019. Improved zero-shot neural machine translation via ignoring spurious correlations. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1258– 1268, Florence, Italy. Association for Computationa…
  11. Katharina Hämmerl, Jindřich Libovický, and Alexander Fraser. 2024. Understanding cross-lingual Alignment—A survey. In Findings of the Association for Computational Linguistics: ACL 2024, pages 10922–10943, Bangkok, Thailand. Association for Computational Linguistics. find
  12. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR).
  13. Carolin Holtermann, Paul Röttger, Timm Dill, and Anne Lauscher. 2024. Evaluating the elementary multilingual capabilities of large language models with MultiQ. In Findings of the Association for Computational Linguistics: ACL 2024, pages 4476–4494, Bangkok, Thailand. Association for Computational Li… find
  14. Chongxuan Huang, Yongshi Ye, Biao Fu, Qifeng Su, and Xiaodong Shi. 2025. From neurons to semantics: Evaluating cross-linguistic alignment capabilities of large language models via neurons alignment. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1:… find
  15. Amir Hossein Kargaran, Ali Modarressi, Nafiseh Nikeghbal, Jana Diesner, François Yvon, and Hinrich Schuetze. 2025. MEXA: Multilingual evaluation of English-centric LLMs via cross-lingual alignment. In Findings of the Association for Computational Linguistics: ACL 2025, pages 27001–27023, Vienna, Aus… find
  16. Chong Li, Shaonan Wang, Jiajun Zhang, and Chengqing Zong. 2024. Improving in-context learning of multilingual generative language models with cross-lingual alignment. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language… find
  17. Xuechen Li, Tianyi Zhang, Yann Dubois, Rohan Taori, Ishaan Gulrajani, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. 2023. Alpacaeval: An automatic evaluator of instruction-following models. https://github.com/tatsu-lab/alpaca_eval. ↗
  18. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. TruthfulQA: Measuring how models mimic human falsehoods. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3214–3252, Dublin, Ireland. Association for Computational Linguistics…
  19. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. find
  20. Yu Meng, Mengzhou Xia, and Danqi Chen. 2024. Simpo: Simple preference optimization with a reference-free reward. In Advances in Neural Information Processing Systems (NeurIPS). find
  21. Richard Yuanzhe Pang, Weizhe Yuan, He He, Kyunghyun Cho, Sainbayar Sukhbaatar, and Jason E Weston. 2024. Iterative reasoning preference optimization. In The Thirty-eighth Annual Conference on Neural Information Processing Systems.
  22. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. In Thirty-seventh Conference on Neural Information Processing Systems.
  23. Leonardo Ranaldi, Giulia Pucci, and Andre Freitas. 2024. Empowering cross-lingual abilities of instruction-tuned large language models by translation-following demonstrations. In Findings of the Association for Computational Linguistics: ACL 2024, pages 7961–7973, Bangkok, Thailand. Association for… find
  24. Jeff Rasley, Samyam Rajbhandari, Olatunji Ruwase, and Yuxiong He. 2020. Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’20, page 3505–3506,… find
  25. Peter J. Rousseeuw. 1987. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20:53–65. find
  26. Rico Sennrich, Jannis Vamvas, and Alireza Mohammadshahi. 2024. Mitigating hallucinations and off-target machine translation with source-contrastive and language-contrastive decoding. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volum…
  27. Shuaijie She, Wei Zou, Shujian Huang, Wenhao Zhu, Xiang Liu, Xiang Geng, and Jiajun Chen. 2024. MAPO: Advancing multilingual reasoning through multilingual-alignment-as-preference optimization. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long…
  28. Tianyi Tang, Wenyang Luo, Haoyang Huang, Dongdong Zhang, Xiaolei Wang, Xin Zhao, Furu Wei, and Ji-Rong Wen. 2024. Language-specific neurons: The key to multilingual capabilities in large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volu…
  29. NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti, and 20 o… find
  30. Khanh-Tung Tran, Barry O’Sullivan, and Hoang Nguyen. 2024. Irish-based large language model with extreme low-resource settings in machine translation. In Proceedings of the Seventh Workshop on Technologies for Machine Translation of Low-Resource Languages (LoResMT 2024), pages 193– 202, Bangkok, Tha… find
  31. Khanh-Tung Tran, Nguyet-Hang Vu, Barry O’Sullivan, and Hoang D. Nguyen. 2025. Disentangling language understanding and reasoning structures in cross-lingual chain-of-thought prompting. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 12200–12206, Suzhou, China. Associa… find
  32. Leandro von Werra, Younes Belkada, Lewis Tunstall, Edward Beeching, Tristan Thrush, Nathan Lambert, Shengyi Huang, Kashif Rasul, and Quentin Gallouédec. 2020. Trl: Transformer reinforcement learning. https://github.com/huggingface/trl. ↗
  33. Chris Wendler, Veniamin Veselovsky, Giovanni Monea, and Robert West. 2024. Do llamas work in English? on the latent language of multilingual transformers. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15366–15394, Bangkok, T…
  34. Zhaofeng Wu, Xinyan Velocity Yu, Dani Yogatama, Jiasen Lu, and Yoon Kim. 2025. The semantic hub hypothesis: Language models share semantic representations across languages and modalities. In The Thirteenth International Conference on Learning Representations. find
  35. Wen Yang, Junhong Wu, Chen Wang, Chengqing Zong, and Jiajun Zhang. 2025a. Implicit cross-lingual rewarding for efficient multilingual preference alignment. In Findings of the Association for Computational Linguistics: ACL 2025, pages 21125–21147, Vienna, Austria. Association for Computational Lingui… find
  36. Wen Yang, Junhong Wu, Chen Wang, Chengqing Zong, and Jiajun Zhang. 2025b. Language imbalance driven rewarding for multilingual self-improving. In The Thirteenth International Conference on Learning Representations. find
  37. Xue Zhang, Yunlong Liang, Fandong Meng, Songming Zhang, Yufeng Chen, Jinan Xu, and Jie Zhou. 2025. Cm-align: Consistency-based multilingual alignment for large language models. Preprint, arXiv:2509.08541. find
  38. Yuanchi Zhang, Yile Wang, Zijun Liu, Shuo Wang, Xiaolong Wang, Peng Li, Maosong Sun, and Yang Liu. 2024b. Enhancing multilingual capabilities of large language models through self-distillation from resource-rich languages. In Proceedings of the 62nd Annual Meeting of the Association for Computationa…
  39. Zhihan Zhang, Dong-Ho Lee, Yuwei Fang, Wenhao Yu, Mengzhao Jia, Meng Jiang, and Francesco Barbieri. 2024c. PLUG: Leveraging pivot language in cross-lingual instruction tuning. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 70…
  40. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. In Proceedings of the 37th International Conference on Neu…
  41. Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho, Xian Li, Sainbayar Sukhbaatar, Jing Xu, and Jason Weston. 2024. Self-rewarding language models. In Proceedings of the 41st International Conference on Machine Learning, ICML’24. JMLR.org. find
  42. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. 2019. HellaSwag: Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791–4800, Florence, Italy. Association for Computational Linguistics. find
  43. Liang Zhang, Qin Jin, Haoyang Huang, Dongdong Zhang, and Furu Wei. 2024a. Respond in my language: Mitigating language inconsistency in response generation based on large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers… find
  44. Xiang Zhang, Senyu Li, Bradley Hauer, Ning Shi, and Grzegorz Kondrak. 2023. Don‘t trust ChatGPT when your question is not in English: A study of multilingual abilities and types of LLMs. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 7915–7927, Singa… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tran-2026-lacomsa-language-consistency-multilingual-self/