Elamur Research⌕
☰
home / papers / 2026 / Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning

Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning

Paper · 2026 · Trapoom Ukarapol, Pakhapoom Sarapat and Nut Chukamphaeng

Summary

English-centric LLMs often slip into the wrong language, usually English, when asked to write in another language. This is called language confusion. The usual fixes are lower temperature, few-shot prompting or multilingual fine-tuning, and they cost either output diversity or extra compute. The authors propose a tuning-free alternative that acts on the logits at decoding time. Tokens whose characters all fall in the target language's Unicode range are identified, along with digits, special symbols and end-of-sequence tokens. A constant α is then added to their logits. This is Language-Aware Token Boosting (LATB). The Adaptive-LATB variant applies the boost only when the top target-language token and the top other-language token are within a threshold β in probability. This lets the model switch languages when it is confident. Tested with Llama3 8B Instruct on XLSUM in eight languages, both methods nearly eliminate confusion, beat the multilingual fine-tune Suzume 8B Multilingual, and keep ROUGE quality. The approach offers a cheap, plug-in way to make English-centric models usable in other languages.

Key points

  • Setup: XLSUM summarization with up to 1,000 test samples per language. Languages cover four high-resource ones (ru, zh, ja, fr) and four medium-resource ones (ko, th, hi, ar).
  • Baselines: a plain prompt, a strict prompt that names the target language, and Suzume 8B Multilingual.
  • Metrics: confusion is measured at token level (Unicode), line level and response level (FastText LID). Summary quality uses ROUGE-1/2/L.
  • Confusion results:
    • Base Llama3 8B Instruct shows about 92–99.6% response-level confusion for every language except French (about 40.6%).
    • With LATB, response-level confusion falls to 0.30% or less for seven languages.
    • Korean remains the hardest case at 10.36% with LATB and 11.45% with Adaptive-LATB, versus 12.36% for Suzume.
  • Theory: adding the same constant to all boosted logits leaves the probability ratios among those tokens unchanged.
  • Speed (vLLM, A100):
    • Base model: 1145.8 tokens/s.
    • LATB: 1189.5 tokens/s, so no real overhead.
    • Adaptive-LATB: 838 tokens/s.
  • Limitations:
    • The method struggles with out-of-vocabulary tokens.
    • Unicode-based identification is weak for scripts that overlap with Latin.
    • Hyperparameters still need tuning.

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

🎓Tsinghua UniversityBeijing · China333 works · 87,019 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 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksMultilingual & cross-lingual NLP · 828 works

Cite

Cite as Trapoom Ukarapol, Pakhapoom Sarapat and Nut Chukamphaeng. 2026. Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-short.40

@inproceedings{ukarapol2026language,
  title = {{Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning}},
  author = {Trapoom Ukarapol, Pakhapoom Sarapat and Nut Chukamphaeng},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-short.40},
  url = {https://doi.org/10.18653/v1/2026.acl-short.40}
}

References

44 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models. Preprint, arXiv:2407.21783.
  2. Muhammad Usman Hadi, Qasem Al Tashi, Abbas Shah, Rizwan Qureshi, Amgad Muneer, Muhammad Irfan, Anas Zafar, Muhammad Bilal Shaikh, Naveed Akhtar, Jia Wu, et al. 2024. Large language models: a comprehensive survey of its applications, challenges, limitations, and future prospects. Authorea Preprints. find
  3. Muhammad Usman Hadi, Rizwan Qureshi, Abbas Shah, Muhammad Irfan, Anas Zafar, Muhammad Bilal Shaikh, Naveed Akhtar, Jia Wu, Seyedali Mirjalili, et al. 2023. A survey on large language models: Applications, challenges, limitations, and practical usage. Authorea Preprints. find
  4. Ahmed Abdelali, Hamdy Mubarak, Shammur Absar Chowdhury, Maram Hasanain, Basel Mousi, Sabri Boughorbel, Yassine El Kheir, Daniel Izham, Fahim Dalvi, Majd Hawasly, et al. 2024. Larabench: Benchmarking arabic ai with large language models. Preprint, arXiv:2305.14982.
  5. Tahmid Hasan, Abhik Bhattacharjee, Md Saiful Islam, Kazi Samin, Yuan-Fang Li, Yong-Bin Kang, M. Sohel Rahman, and Rifat Shahriyar. 2021. Xl-sum: Large-scale multilingual abstractive summarization for 44 languages. Preprint, arXiv:2106.13822. find
  6. Arav Agarwal, Karthik Mittal, Aidan Doyle, Pragnya Sridhar, Zipiao Wan, Jacob Arthur Doughty, Jaromir Savelka, and Majd Sakr. 2024. Understanding the role of temperature in diverse question generation by gpt-4. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 2, SI… find
  7. Zhiwei He, Tian Liang, Wenxiang Jiao, Zhuosheng Zhang, Yujiu Yang, Rui Wang, Zhaopeng Tu, Shuming Shi, and Xing Wang. 2023. Exploring human-like translation strategy with large language models. Preprint, arXiv:2305.04118.
  8. Carolin Holtermann, Paul Röttger, Timm Dill, and Anne Lauscher. 2024. Evaluating the elementary multilingual capabilities of large language models with multiq. Preprint, arXiv:2403.03814. find
  9. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen technical report. Preprint, arXiv:2309.16609.
  10. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. 2022. Palm: Scaling language modeling with pathways. Preprint, arXiv:2204.02311.
  11. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2022. Scaling instruction-finetuned language models. Preprint, arXiv:2210.11416. find
  12. Simone Conia, Min Li, Daniel Lee, Umar Farooq Minhas, Ihab Ilyas, and Yunyao Li. 2023. Increasing coverage and precision of textual information in multilingual knowledge graphs. Preprint, arXiv:2311.15781. find
  13. Peter Devine. 2024. Tagengo: A multilingual chat dataset. Preprint, arXiv:2405.12612. find
  14. Team GLM, :, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. 2024. Chatglm: A family of large language models from glm-130b to glm-4 all tools. Preprint, arXiv:2406.12793.
  15. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. Preprint, arXiv:2310.06825. find
  16. Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al. 2024. Mixtral of experts. Preprint, arXiv:2401.04088. find
  17. Armand Joulin, Edouard Grave, Piotr Bojanowski, Matthijs Douze, Hérve Jégou, and Tomas Mikolov. 2016a. Fasttext.zip: Compressing text classification models. arXiv preprint arXiv:1612.03651.
  18. Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. 2016b. Bag of tricks for efficient text classification. arXiv preprint arXiv:1607.01759. find
  19. Viet Dac Lai, Nghia Trung Ngo, Amir Pouran Ben Veyseh, Hieu Man, Franck Dernoncourt, Trung Bui, and Thien Huu Nguyen. 2023a. Chatgpt beyond english: Towards a comprehensive evaluation of large language models in multilingual learning. Preprint, arXiv:2304.05613.
  20. Viet Dac Lai, Chien Van Nguyen, Nghia Trung Ngo, Thuat Nguyen, Franck Dernoncourt, Ryan A. Rossi, and Thien Huu Nguyen. 2023b. Okapi: Instruction-tuned large language models in multiple languages with reinforcement learning from human feedback. Preprint, arXiv:2307.16039.
  21. Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. 2020. Gshard: Scaling giant models with conditional computation and automatic sharding. Preprint, arXiv:2006.16668. find
  22. Tianjian Li and Kenton Murray. 2023. Why does zero-shot cross-lingual generation fail? an explanation and a solution. Preprint, arXiv:2305.17325. find
  23. Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics.
  24. Hongyuan Lu, Haoran Yang, Haoyang Huang, Dongdong Zhang, Wai Lam, and Furu Wei. 2024. Chain-of-dictionary prompting elicits translation in large language models. Preprint, arXiv:2305.06575.
  25. Kelly Marchisio, Wei-Yin Ko, Alexandre Bérard, Théo Dehaze, and Sebastian Ruder. 2024. Understanding and mitigating language confusion in llms. Preprint, arXiv:2406.20052.
  26. Niklas Muennighoff, Thomas Wang, Lintang Sutawika, Adam Roberts, Stella Biderman, Teven Le Scao, M Saiful Bari, Sheng Shen, Zheng-Xin Yong, Hailey Schoelkopf, et al. 2023. Crosslingual generalization through multitask finetuning. Preprint, arXiv:2211.01786.
  27. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, et al. 2024. Gpt-4 technical report. Preprint, arXiv:2303.08774.
  28. Kunat Pipatanakul, Phatrasek Jirabovonvisut, Potsawee Manakul, Sittipong Sripaisarnmongkol, Ruangsak Patomwong, Pathomporn Chokchainant, and Kasima Tharnpipitchai. 2023. Typhoon: Thai large language models. Preprint, arXiv:2312.13951. — [arXiv:2312.13951](https://arxiv.org/abs/2312.13951) find
  29. Zhi Rui Tam, Cheng-Kuang Wu, Yu Ying Chiu, Chieh-Yen Lin, Yun-Nung Chen, and Hung yi Lee. 2025. Language matters: How do multilingual input and reasoning paths affect large reasoning models? Preprint, arXiv:2505.17407.
  30. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. Preprint, arXiv:2307.09288.
  31. Tu Vu, Aditya Barua, Brian Lester, Daniel Cer, Mohit Iyyer, and Noah Constant. 2022. Overcoming catastrophic forgetting in zero-shot cross-lingual generation. Preprint, arXiv:2205.12647. find
  32. Jiaan Wang, Fandong Meng, Yunlong Liang, Tingyi Zhang, Jiarong Xu, Zhixu Li, and Jie Zhou. 2023. Understanding translationese in cross-lingual summarization. Preprint, arXiv:2212.07220. find
  33. Mingyang Wang, Lukas Lange, Heike Adel, Yunpu Ma, Jannik Strötgen, and Hinrich Schütze. 2025. Language mixing in reasoning language models: Patterns, impact, and internal causes. Preprint, arXiv:2505.14815.
  34. Xiangpeng Wei, Haoran Wei, Huan Lin, Tianhao Li, Pei Zhang, Xingzhang Ren, Mei Li, Yu Wan, Zhiwei Cao, Binbin Xie, et al. 2023. Polylm: An open source polyglot large language model. Preprint, arXiv:2307.06018. find
  35. Andrea W Wen-Yi and David Mimno. 2023. Hyperpoly-glot LLMs: Cross-lingual interpretability in token embeddings. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 1124–1131, Singapore. Association for Computational Linguistics. find
  36. Genta Indra Winata, Alham Fikri Aji, Zheng-Xin Yong, and Thamar Solorio. 2023. The decades progress on code-switching research in nlp: A systematic survey on trends and challenges. Preprint, arXiv:2212.09660. find
  37. Ratish Puduppully, Anoop Kunchukuttan, Raj Dabre, Ai Ti Aw, and Nancy F. Chen. 2023. Decomposed prompting for machine translation between related languages using large language models. Preprint, arXiv:2305.13085. find
  38. Libo Qin, Qiguang Chen, Yuhang Zhou, Zhi Chen, Yinghui Li, Lizi Liao, Min Li, Wanxiang Che, and Philip S. Yu. 2024. Multilingual large language model: A survey of resources, taxonomy and frontiers. Preprint, arXiv:2404.04925. find
  39. Matthew Renze and Erhan Guven. 2024. The effect of sampling temperature on problem solving in large language models. Preprint, arXiv:2402.05201.
  40. Andy Rosenbaum, Saleh Soltan, Wael Hamza, Yannick Versley, and Markus Boese. 2022. Linguist: Language model instruction tuning to generate annotated utterances for intent classification and slot tagging. Preprint, arXiv:2209.09900. find
  41. BigScience Workshop, :, Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilić, Daniel Hesslow, Roman Castagné, Alexandra Sasha Luccioni, François Yvon, et al. 2023. Bloom: A 176b-parameter open-access multilingual language model. Preprint, arXiv:2211.05100. find
  42. Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mt5: A massively multilingual pre-trained text-to-text transformer. Preprint, arXiv:2010.11934.
  43. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jin… — Qwen3 Technical Report
  44. Min Zhang, Limin Liu, Zhao Yanqing, Xiaosong Qiao, Su Chang, Xiaofeng Zhao, Junhao Zhu, Ming Zhu, Song Peng, Yinglu Li, et al. 2023. Leveraging multilingual knowledge graph to boost domain-specific entity translation of ChatGPT. In Proceedings of Machine Translation Summit XIX, Vol. 2: Users Track,… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ukarapol-2026-language-aware-token-boosting-llm/