Elamur Research
☰
home / papers / 2023 / NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark

NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/2023.findings-emnlp.722.pdf

Summary

This position paper argues that benchmark-based NLP evaluation is unreliable when large language models have already seen the test data. Models are trained on huge web crawls, so it is often unclear whether a benchmark's test split was part of their training data. A contaminated model scores higher than it should. This skews comparisons between LLMs and, more seriously, can lead to published scientific claims that are wrong while correct hypotheses get rejected. The authors, from HiTZ (UPV/EHU) and Cohere, define three types of contamination: guideline, raw text and annotation. They also show where it can enter a model's development: pre-training, instruction fine-tuning, and retraining on user inputs after deployment. For open models, they recommend data-search tools and reporting how much of a benchmark appears in the training data. For closed models, they propose testing whether the model can reproduce the benchmark from memory. They call for automatic detection methods, a shared registry of contamination cases, and review procedures that flag affected papers. This matters because reported zero-shot or few-shot gains may come from leaked test data rather than real ability.

Key points

  • Annotation contamination: if the model has seen the labels of the test split, the experiment is fully invalid. If it has seen only the train or dev labels, claims of zero-shot or few-shot performance no longer hold, but comparisons with models trained on those same splits are still fair.
  • Documented cases: the C4 corpus contains test splits of several benchmarks. A filtering bug caused GPT-3 to train on benchmark data. The GPT-4 report admits that BIG-bench data leaked into training and that parts of the MATH and GSM-8K training sets were used on purpose. The ROOTS Search Tool showed that BLOOM was trained on XNLI data.
  • CoNLL2003: ChatGPT can reproduce parts of this NER benchmark. Appendix A shows memorization evidence for three LLMs, which casts doubt on several preprints and an ACL 2023 paper that evaluated GPT-3 and Codex on it.
  • Proposed metric for closed models: an example counts as extractable if the model generates it when prompted with the k examples that come before it. A model's contamination degree is the share of a dataset's examples that are extractable. The authors leave open whether a model that has not memorized a benchmark still gains from having trained on it.
  • Resources: the LM Contamination Index registry (hitz-zentroa.github.io/lm-contamination/) and the 1st Workshop on Data Contamination.

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

Fields: Mathematics & statistics

Cited by

164 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 162 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.findings-emnlp.722 · arXiv 2310.18018

@inproceedings{sainz2023nlp,
  title = {{NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark}},
  author = {Sainz, Oscar and Campos, Jon and García-Ferrero, Iker and Etxaniz, Julen and Lacalle, Oier Lopez de and Agirre, Eneko},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.findings-emnlp.722},
  eprint = {2310.18018},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2023.findings-emnlp.722}
}

References

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

  1. Stephen Bach, Victor Sanh, Zheng Xin Yong, Albert Webson, Colin Raffel, Nihal V. Nayak, Abheesht Sharma, Taewoon Kim, M Saiful Bari, Thibault Fevry, Zaid Alyafeai, Manan Dey, Andrea Santilli, Zhiqing Sun, Srulik Ben-david, Canwen Xu, Gunjan Chhablani, Han Wang, Jason Fries, Maged Al-shaibani, Shanya… find
  2. Abeba Birhane, Vinay Uday Prabhu, and Emmanuel Kahembwe. 2021. Multimodal datasets: misogyny, pornography, and malignant stereotypes. https://conda-workshop.github.io ↗
  3. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… find
  4. Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tramer, and Chiyuan Zhang. 2023. Quantifying memorization across neural language models. In The Eleventh International Conference on Learning Representations. find
  5. Nicholas Carlini, Florian Tramèr, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Úlfar Erlingsson, Alina Oprea, and Colin Raffel. 2021. Extracting training data from large language models. In 30th USENIX Security Symposium (USENIX Security 21)… find our card
  6. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhai… find our card
  7. Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality. find
  8. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168. find our card
  9. Alexis Conneau, Ruty Rinott, Guillaume Lample, Adina Williams, Samuel Bowman, Holger Schwenk, and Veselin Stoyanov. 2018. XNLI: Evaluating cross-lingual sentence representations. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2475–2485, Brussels, Bel… find
  10. Jesse Dodge, Maarten Sap, Ana Marasović, William Agnew, Gabriel Ilharco, Dirk Groeneveld, Margaret Mitchell, and Matt Gardner. 2021. Documenting large webtext corpora: A case study on the colossal clean crawled corpus. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language P… find
  11. Nan Du, Yanping Huang, Andrew M. Dai, Simon Tong, Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun, Yanqi Zhou, Adams Wei Yu, Orhan Firat, Barret Zoph, Liam Fedus, Maarten Bosma, Zongwei Zhou, Tao Wang, Yu Emma Wang, Kellie Webster, Marie Pellat, Kevin Robinson, Kathy Meier-Hellstern, Toju Duke, Lucas Di… find our card
  12. Aparna Elangovan, Jiayuan He, and Karin Verspoor. 2021. Memorization vs. generalization : Quantifying data leakage in NLP performance evaluation. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 1325–1335, Online. Asso… find
  13. Besnik Fetahu, Sudipta Kar, Zhiyu Chen, Oleg Rokhlenko, and Shervin Malmasi. 2023. SemEval-2023 Task 2: Fine-grained Multilingual Named Entity Recognition (MultiCoNER 2). In Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023). Association for Computational Linguistic… find
  14. Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A. Smith. 2020. RealToxicityPrompts: Evaluating neural toxic degeneration in language models. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3356–3369, Online. Association for Computational Linguisti… find our card
  15. Sophie Groenwold, Lily Ou, Aesha Parekh, Samhita Honnavalli, Sharon Levy, Diba Mirza, and William Yang Wang. 2020. Investigating African-American Vernacular English in transformer-based text generation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)… find
  16. Yuxian Gu, Li Dong, Furu Wei, and Minlie Huang. 2023. Knowledge distillation of large language models. find
  17. Ridong Han, Tao Peng, Chaohao Yang, Benyou Wang, Lu Liu, and Xiang Wan. 2023. Is information extraction solved by chatgpt? an analysis of performance, evaluation criteria, robustness and errors. find
  18. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. arXiv preprint arXiv:2103.03874. find our card
  19. Tom Henighan, Jared Kaplan, Mor Katz, Mark Chen, Christopher Hesse, Jacob Jackson, Heewoo Jun, Tom B. Brown, Prafulla Dhariwal, Scott Gray, Chris Hallacy, Benjamin Mann, Alec Radford, Aditya Ramesh, Nick Ryder, Daniel M. Ziegler, John Schulman, Dario Amodei, and Sam McCandlish. 2020. Scaling laws fo… find
  20. Karl Moritz Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. 2015. Teaching machines to read and comprehend. In NIPS, pages 1693–1701. find our card
  21. Or Honovich, Thomas Scialom, Omer Levy, and Timo Schick. 2022. Unnatural instructions: Tuning language models with (almost) no human labor. arXiv preprint arXiv:2212.09689. find our card
  22. Srinivasan Iyer, Xi Victoria Lin, Ramakanth Pasunuru, Todor Mihaylov, Daniel Simig, Ping Yu, Kurt Shuster, Tianlu Wang, Qing Liu, Punit Singh Koura, Xian Li, Brian O’Horo, Gabriel Pereyra, Jeff Wang, Christopher Dewan, Asli Celikyilmaz, Luke Zettlemoyer, and Ves Stoyanov. 2023. Opt-iml: Scaling lang… find our card
  23. Alon Jacovi, Avi Caciularu, Omer Goldman, and Yoav Goldberg. 2023. Stop uploading test data in plain text: Practical strategies for mitigating data contamination by evaluation benchmarks. find
  24. Wenxiang Jiao, Wenxuan Wang, Jen tse Huang, Xing Wang, and Zhaopeng Tu. 2023. Is chatgpt a good translator? yes with gpt-4 as the engine. find our card
  25. Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. Scaling laws for neural language models. find
  26. Julia Kreutzer, Isaac Caswell, Lisa Wang, Ahsan Wahab, Daan van Esch, Nasanbayar Ulzii-Orshikh, Allahsera Tapo, Nishant Subramani, Artem Sokolov, Claytone Sikasote, Monang Setyawan, Supheakmungkol Sarin, Sokhar Samb, Benoît Sagot, Clara Rivera, Annette Rios, Isabel Papadimitriou, Salomey Osei, Pedro… find our card
  27. Katherine Lee, Daphne Ippolito, Andrew Nystrom, Chiyuan Zhang, Douglas Eck, Chris Callison-Burch, and Nicholas Carlini. 2022. Deduplicating training data makes language models better. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),… find
  28. Bo Li, Gexiang Fang, Yang Yang, Quansen Wang, Wei Ye, Wen Zhao, and Shikun Zhang. 2023a. Evaluating chatgpt’s information extraction capabilities: An assessment of performance, explainability, calibration, and faithfulness. find
  29. Peng Li, Tianxiang Sun, Qiong Tang, Hang Yan, Yuanbin Wu, Xuanjing Huang, and Xipeng Qiu. 2023b. Codeie: Large code generation models are better few-shot information extractors. In Proceedings of the 61th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Toront… find
  30. Tao Li, Daniel Khashabi, Tushar Khot, Ashish Sabharwal, and Vivek Srikumar. 2020. UNQOVERing stereotyping biases via underspecified questions. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3475–3489, Online. Association for Computational Linguistics. find
  31. Shayne Longpre, Le Hou, Tu Vu, Albert Webson, Hyung Won Chung, Yi Tay, Denny Zhou, Quoc V. Le, Barret Zoph, Jason Wei, and Adam Roberts. 2023. The flan collection: Designing data and methods for effective instruction tuning. find our card
  32. Ziyang Luo, Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, and Daxin Jiang. 2023. Wizardcoder: Empowering code large language models with evol-instruct. find
  33. Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts. 2011. Learning word vectors for sentiment analysis. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 142–150, Portland, Oregon… find our card
  34. Inbal Magar and Roy Schwartz. 2022. Data contamination: From memorization to exploitation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 157–165, Dublin, Ireland. Association for Computational Linguistics. find
  35. Marc Marone and Benjamin Van Durme. 2023. Data portraits: Recording foundation model training data. find
  36. Bonan Min, Hayley Ross, Elior Sulem, Amir Pouran Ben Veyseh, Thien Huu Nguyen, Oscar Sainz, Eneko Agirre, Ilana Heinz, and Dan Roth. 2021. Recent advances in natural language processing via large pre-trained language models: A survey. find our card
  37. OpenAI. 2023. Gpt-4 technical report. find our card
  38. Aleksandra Piktus, Christopher Akiki, Paulo Villegas, Hugo Laurençon, Gérard Dupont, Alexandra Sasha Luccioni, Yacine Jernite, and Anna Rogers. 2023. The roots search tool: Data transparency for llms. find
  39. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67. find
  40. Oscar Sainz, Jon Ander Campos, Iker García-Ferrero, Julen Etxaniz, and Eneko Agirre. 2023. Did chatgpt cheat on your test? find
  41. Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, Agnieszka Kluska, Aitor Lewkowycz, Akshat Agarwal, Alethea Power, Alex Ray, Alex Warstadt, Alexander W. Kocurek, Ali Safaya, Ali Tazarv, Al… find our card
  42. Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. 2023. Stanford alpaca: An instruction-following llama model. https:// github.com/tatsu-lab/stanford\_alpaca. find
  43. Erik F. Tjong Kim Sang and Fien De Meulder. 2003. Introduction to the CoNLL-2003 shared task: Language-independent named entity recognition. In Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003, pages 142– 147. find
  44. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023a. Llama: Open and efficient foundation language models. find
  45. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton-Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… find
  46. Veniamin Veselovsky, Manoel Horta Ribeiro, and Robert West. 2023. Artificial artificial artificial intelligence: Crowd workers widely use large language models for text production tasks. find
  47. Christopher Walker, Stephanie Strassel, Julie Medero, and Kazuaki Maeda. 2006. find
  48. Ace 2005 multilingual training corpus. Linguistic Data Consortium, Philadelphia, 57:45. find
  49. Alex Wang, Yada Pruksachatkun, Nikita Nangia, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. 2020. Superglue: A stickier benchmark for general-purpose language understanding systems. find our card
  50. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. 2018. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 353–355,… find our card
  51. Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A Smith, Daniel Khashabi, and Hannaneh Hajishirzi. 2022a. Self-instruct: Aligning language model with self generated instructions. arXiv preprint arXiv:2212.10560. find
  52. Yizhong Wang, Swaroop Mishra, Pegah Alipoormo-labashi, Yeganeh Kordi, Amirreza Mirzaei, Atharva Naik, Arjun Ashok, Arut Selvan Dhanasekaran, Anjana Arunkumar, David Stap, Eshaan Pathak, Giannis Karamanolakis, Haizhi Lai, Ishan Purohit, Ishani Mondal, Jacob Anderson, Kirby Kuznia, Krima Doshi, Kuntal… find
  53. Jason Wei, Maarten Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V Le. 2022. Finetuned language models are zero-shot learners. In International Conference on Learning Representations. find
  54. Xiang Wei, Xingyu Cui, Ning Cheng, Xiaobin Wang, Xin Zhang, Shen Huang, Pengjun Xie, Jinan Xu, Yufeng Chen, Meishan Zhang, Yong Jiang, and Wenjuan Han. 2023. Zero-shot information extraction via chatting with chatgpt. find
  55. BigScience Workshop, :, Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilić, Daniel Hesslow, Roman Castagné, Alexandra Sasha Luccioni, François Yvon, Matthias Gallé, Jonathan Tow, Alexander M. Rush, Stella Biderman, Albert Webson, Pawan Sasanka Ammanamanchi, Thomas Wang, Benoît… find