Elamur Research⌕
☰
home / papers / 2024 / The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts

The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts

Paper · 2024 · Lingfeng Shen, Weiting Tan, Sihao Chen et al.

Summary

This paper asks whether the safety behaviour of large language models holds up when the same harmful request is posed in languages with little training data. The authors build a simple translation attack. They take English malicious prompts from Zou et al. (2023), translate them with NLLB-1.3B into 9 high-resource and 10 low-resource languages, query the model, and translate the replies back to English. GPT-4 then labels each reply as harmful, harmless, or irrelevant. For GPT-4 itself, they find two weaknesses in low-resource languages: it produces harmful content far more often, and it more often fails to understand the instruction at all. They then align LLaMa2 on translated HH-RLHF data using multilingual SFT (xSFT) and multilingual RLHF (xRLHF). Both methods help in well-resourced languages but barely help in low-resourced ones. The authors therefore trace the problem to limited pre-training exposure rather than to alignment. This matters for languages with fewer resources, where models may be both less safe and less useful.

Key points

  • The paper defines two metrics. HARMFUL RATE is the share of on-topic replies that are harmful. FOLLOWING RATE is the share of replies that are not irrelevant.
  • For GPT-4, low-resource languages had harmful rates of roughly 25–38%, compared with 0–2% for high-resource ones. Following rates were about 65–82% versus 98–100%.
  • Urdu, Igbo, Maori, Armenian, Kamba, Hausa, Halh Mongolian, Javanese and Luo are among the languages in the low-resource group. The high-resource set covers Chinese, Russian, Spanish, French, German, Turkish and others.
  • On LLaMa2, xSFT reduced harmfulness more than xRLHF. Average harmful rates were 57.4/70.6 (high/low) for xSFT versus 66.0/78.0 for xRLHF.
  • The multilingual reward model (X-RM) reached about 63% accuracy on high-resource languages but only about 49% on low-resource ones, which is close to chance. Adding CONTRAST INSTRUCTION did not change this.

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

🎓University of PennsylvaniaPhiladelphia · United States170 works · 60,739 citations🎓Johns Hopkins UniversityBaltimore · United States187 works · 92,194 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 1,343 worksMultilingual & cross-lingual NLP · 467 worksBias, fairness & safety · 209 works

Cite

Cite as Lingfeng Shen, Weiting Tan, Sihao Chen, Yunmo Chen, Jingyu Zhang, Haoran Xu, Boyuan Zheng, Philipp Koehn and Daniel Khashabi. 2024. The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts. Findings of the Association for Computational Linguistics: ACL 2024.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2024.findings-acl.156

@inproceedings{shen2024language,
  title = {{The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts}},
  author = {Lingfeng Shen, Weiting Tan, Sihao Chen, Yunmo Chen, Jingyu Zhang, Haoran Xu, Boyuan Zheng, Philipp Koehn and Daniel Khashabi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2024},
  year = {2024},
  doi = {10.18653/v1/2024.findings-acl.156},
  url = {https://doi.org/10.18653/v1/2024.findings-acl.156}
}

References

46 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ayush Agrawal, Mirac Suzgun, Lester Mackey, and Adam Tauman Kalai. 2023. Do language models know when they’re hallucinating references? find
  2. Jordi Armengol-Estapé, Ona de Gibert Bonet, and Maite Melero. 2022. On the multilingual capabilities of very large-scale English language models. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 3056–3068, Marseille, France. European Language Resources Associatio… find
  3. Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. 2022a. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862. find
  4. Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. 2022b. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073.
  5. Damian Blasi, Antonios Anastasopoulos, and Graham Neubig. 2022. Systematic inequalities in language technology performance across the world’s languages. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5486–5505, Dublin, Irelan…
  6. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam Shazeer, Vinodkumar Prabhakaran, Emi…
  7. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. find
  8. Yue Deng, Wenxuan Zhang, Sinno Jialin Pan, and Lidong Bing. 2024. Multilingual jailbreak challenges in large language models. In The Thirteenth International Conference on Learning Representations.
  9. Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and Jason Weston. 2023. Chain-of-verification reduces hallucination in large language models.
  10. Nouha Dziri, Sivan Milton, Mo Yu, Osmar Zaiane, and Siva Reddy. 2022. On the origin of hallucinations in conversational models: Is it the datasets or the models? In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techn…
  11. Deep Ganguli, Amanda Askell, Nicholas Schiefer, Thomas Liao, Kamilė Lukošiūtė, Anna Chen, Anna Goldie, Azalia Mirhoseini, Catherine Olsson, Danny Hernandez, et al. 2023. The capacity for moral self-correction in large language models. arXiv preprint arXiv:2302.07459.
  12. Deep Ganguli, Liane Lovitt, Jackson Kernion, Amanda Askell, Yuntao Bai, Saurav Kadavath, Ben Mann, Ethan Perez, Nicholas Schiefer, Kamal Ndousse, Andy Jones, Sam Bowman, Anna Chen, Tom Conerly, Nova DasSarma, Dawn Drain, Nelson Elhage, Sheer El-Showk, Stanislav Fort, Zac Hatfield-Dodds, Tom Henighan…
  13. Dongyoung Go, Tomasz Korbak, Germán Kruszewski, Jos Rozen, Nahyeon Ryu, and Marc Dymetman. 2023. Aligning language models with preferences through f-divergence minimization. arXiv preprint arXiv:2302.08215. find
  14. Donald Joseph Hejna III and Dorsa Sadigh. 2023. Few-shot preference learning for human-in-the-loop RL. find
  15. In Conference on Robot Learning (CoRL), pages 2014–2025. find
  16. Edward J Hu, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. 2021. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations (ICLR).
  17. Karthikeyan K, Zihan Wang, Stephen Mayhew, and Dan Roth. 2020. Cross-lingual ability of multilingual bert: An empirical study. find
  18. Katikapalli Subramanyam Kalyan, Ajit Rajasekharan, and Sivanesan Sangeetha. 2021. Ammus : A survey of transformer-based pretrained models in natural language processing. find
  19. Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Hannaneh Hajishirzi. 2020. UnifiedQA: Crossing Format Boundaries With a Single QA System. In Conference on Empirical Methods in Natural Language Processing (EMNLP) - Findings.
  20. Hyunwoo Kim, Youngjae Yu, Liwei Jiang, Ximing Lu, Daniel Khashabi, Gunhee Kim, Yejin Choi, and Maarten Sap. 2022. Prosocialdialog: A prosocial backbone for conversational agents. In Conference on Empirical Methods in Natural Language Processing (EMNLP).
  21. Tomasz Korbak, Kejian Shi, Angelica Chen, Rasika Bhalerao, Christopher L Buckley, Jason Phang, Samuel R Bowman, and Ethan Perez. 2023. Pre-training language models with human preferences. arXiv preprint arXiv:2302.08582.
  22. Haonan Li, Fajri Koto, Minghao Wu, Alham Fikri Aji, and Timothy Baldwin. 2023a. Bactrian-x : A multilingual replicable instruction-following model with low-rank adaptation. find
  23. Haoran Li, Dadi Guo, Wei Fan, Mingshi Xu, Jie Huang, Fanpu Meng, and Yangqiu Song. 2023b. Multi-step jailbreaking privacy attacks on chatgpt.
  24. Lei Li, Yuwei Yin, Shicheng Li, Liang Chen, Peiyi Wang, Shuhuai Ren, Mukai Li, Yazheng Yang, Jingjing Xu, Xu Sun, Lingpeng Kong, and Qi Liu. 2023c. M 3 it: A large-scale dataset towards multimodal multilingual instruction tuning. find
  25. Ximing Lu, Sean Welleck, Jack Hessel, Liwei Jiang, Lianhui Qin, Peter West, Prithviraj Ammanabrolu, and Yejin Choi. 2022. Quark: controllable text generation with reinforced unlearning. Advances in neural information processing systems, 35:27591– 27609.
  26. Marta R NLLB Team, Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, et al. 2022. No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672. find
  27. OpenAI. 2023. GPT-4 technical report. abs/2303.08774.
  28. CoRR, Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training Language Models to Follow Instructions with Human Feedback. In Advances in Neural Information Processing Systems (NeurIPS).
  29. Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilić, Daniel Hesslow, Roman Castagné, Alexandra Sasha Luccioni, François Yvon, Matthias Gallé, et al. 2022. Bloom: A 176b-parameter open-access multilingual language model. arXiv preprint arXiv:2211.05100. find
  30. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. find
  31. Lingfeng Shen, Sihao Chen, Linfeng Song, Lifeng Jin, Baolin Peng, Haitao Mi, Daniel Khashabi, and Dong Yu. 2023a. The trickle-down impact of reward (in-) consistency on rlhf. arXiv preprint arXiv:2309.16155. find
  32. Xinyue Shen, Zeyuan Chen, Michael Backes, Yun Shen, and Yang Zhang. 2023b. ”Do Anything Now”: Characterizing and evaluating in-the-wild jailbreak prompts on large language models. arXiv preprint arXiv:2308.03825.
  33. Xinyue Shen, Zeyuan Chen, Michael Backes, Yun Shen, and Yang Zhang. 2023c. ”do anything now”: Characterizing and evaluating in-the-wild jailbreak prompts on large language models.
  34. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. LLaMA: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971.
  35. Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi, Yeganeh Kordi, Amirreza Mirzaei, Atharva Naik, Arjun Ashok, Arut Selvan Dhanasekaran, Anjana Arunkumar, David Stap, et al. 2022. Super-naturalinstructions: Generalization via declarative instructions on 1600+ nlp tasks. In Proceedings of the 2022… find
  36. Yuxia Wang, Haonan Li, Xudong Han, Preslav Nakov, and Timothy Baldwin. 2023. Do-not-answer: A dataset for evaluating safeguards in llms.
  37. Alexander Wei, Nika Haghtalab, and Jacob Steinhardt. 2023. Jailbroken: How does llm safety training fail? arXiv preprint arXiv:2307.02483.
  38. Yotam Wolf, Noam Wies, Oshri Avnery, Yoav Levine, and Amnon Shashua. 2023. Fundamental limitations of alignment in large language models.
  39. Shijie Wu and Mark Dredze. 2020. Are all languages created equal in multilingual BERT? In Proceedings of the 5th Workshop on Representation Learning for NLP, pages 120–130, Online. Association for Computational Linguistics.
  40. Haoran Xu, Young Jin Kim, Amr Sharaf, and Hany Hassan Awadalla. 2023. A paradigm shift in machine translation: Boosting translation performance of large language models. arXiv preprint arXiv:2309.11674. find
  41. Haoran Xu, Amr Sharaf, Yunmo Chen, Weiting Tan, Lingfeng Shen, Benjamin Van Durme, Kenton Murray, and Young Jin Kim. 2024. Contrastive preference optimization: Pushing the boundaries of llm performance in machine translation.
  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. In Conference of the North American Chapter of the Association for Computational Linguistics (NAACL).
  43. Zheng-Xin Yong, Cristina Menghini, and Stephen H. Bach. 2023. Low-resource languages jailbreak gpt-4.
  44. Shaolei Zhang, Qingkai Fang, Zhuocheng Zhang, Zhengrui Ma, Yan Zhou, Langlin Huang, Mengyu Bu, Shangtong Gui, Yunji Chen, Xilin Chen, and Yang Feng. 2023. Bayling: Bridging cross-lingual alignment and instruction following through interactive translation for large language models. find
  45. Jieyu Zhao, Daniel Khashabi, Tushar Khot, Ashish Sabharwal, and Kai-Wei Chang. 2021. Ethical-advice taker: Do language models understand natural language interventions? In Annual Meeting of the Association for Computational Linguistics (ACL) - Findings. find
  46. Andy Zou, Zifan Wang, J Zico Kolter, and Matt Fredrikson. 2023. Universal and transferable adversarial attacks on aligned language models. arXiv preprint arXiv:2307.15043.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shen-2024-language-barrier-dissecting-safety-challenges/