Elamur Research⌕
☰
home / papers / 2026 / Compact Example-Based Explanations for Language Models

Compact Example-Based Explanations for Language Models

Paper · 2026 · Loris Schoenegger and Benjamin Roth

Summary

Influence estimation methods score how much each training document shaped a language model's output, but people can only inspect a handful of examples. Explanation systems therefore need a selection step, which earlier evaluations have mostly overlooked. Schoenegger and Roth address this gap with a selection relevance score. It treats explanation as gradient reconstruction: a chosen set of k training examples scores well if a linear combination of their loss gradients closely approximates the test instance's gradient. The score judges examples jointly rather than one at a time, needs no retraining, and works with any task or estimator. The authors validated it by correlating it with two fine-tuning-based measures of whether training on a selection reinforces the original prediction. Using the score, they found that common strategies, including taking the top-ranked influential examples, often do worse than random picks, because those examples tend to be global outliers or redundant. They propose a strategy that weighs influence against representativeness. The broader point is that explanation quality depends on how examples are selected, not only on how accurate the influence estimates are.

Key points

  • The benchmark covers 3 influence estimation methods (including DataInf, LESS and a BM25 similarity baseline) and 4 selection strategies, with budgets of k = 1, 5, 10 and 25.
  • BM25-based selection performed strongly, while selections built on the gradient-based estimators scored low overall. The authors note that the "most helpful" examples by influence are not necessarily the most similar ones.
  • Selections with higher relevance scores supported the model's outputs more strongly in fine-tuning experiments than random selections did.
  • To keep computation feasible, gradients were restricted to LoRA layers and the experiments used models of 0.5–1B parameters on instruction fine-tuning data.
  • The authors stress that relevance is a necessary condition for faithfulness but does not guarantee it. Code is available at doi.org/10.5281/zenodo.19483839.

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 ViennaVienna · Austria16 works · 38,729 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.Language models & pretraining · 3,320 works

Cite

Cite as Loris Schoenegger and Benjamin Roth. 2026. Compact Example-Based Explanations for Language Models. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1214

@inproceedings{schoenegger2026compact,
  title = {{Compact Example-Based Explanations for Language Models}},
  author = {Loris Schoenegger and Benjamin Roth},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1214},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1214}
}

References

29 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Juhan Bae, Nathan Ng, Alston Lo, Marzyeh Ghassemi, and Roger B. Grosse. 2022. If Influence Functions are the Answer, Then What is the Question? In Advances in Neural Information Processing Systems, volume 35, pages 17953–17967. find
  2. Elnaz Barshan, Marc-Etienne Brunet, and Gintare Karolina Dziugaite. 2020. RelatIF: Identifying Explanatory Training Samples via Relative Influence. In Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, pages 1899–1909. PMLR. find
  3. Samyadeep Basu, Phil Pope, and Soheil Feizi. 2021. Influence Functions in Deep Learning Are Fragile. In 9th International Conference on Learning Representations. find
  4. Umang Bhatt, Isabel Chien, Muhammad Bilal Zafar, and Adrian Weller. 2021. DIVINE: Diverse Influential Training Points for Data Visualization and Model Refinement. Preprint, arXiv:2107.05978. — [arXiv:2107.05978](https://arxiv.org/abs/2107.05978) find
  5. Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Franco Turini, Fosca Giannotti, and Dino Pedreschi. 2018. A Survey of Methods for Explaining Black Box Models. ACM Comput. Surv., 51(5):93:1–93:42. find
  6. Zayd Hammoudeh and Daniel Lowd. 2022. Identifying a Training-Set Attack’s Target Using Renormalized Influence Estimation. In Proceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security, CCS ’22, pages 1367–1381, New York, NY, USA. Association for Computing Machinery. find
  7. Kazuaki Hanawa, Sho Yokoi, Satoshi Hara, and Kentaro Inui. 2021. Evaluation of Similarity-based Explanations. In 9th International Conference on Learning Representations. find
  8. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. In The Tenth International Conference on Learning Representations.
  9. Sang Keun Choe, Hwijeen Ahn, Juhan Bae, Kewen Zhao, Youngseog Chung, Adithya Pratapa, Willie Neiswanger, Emma Strubell, Teruko Mitamura, Jeff Schneider, Eduard Hovy, Roger Baker Grosse, and Eric P. Xing. 2025. What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions. In The… find
  10. Karthikeyan K and Anders Søgaard. 2021. Revisiting Methods for Finding Influential Examples. Preprint, arXiv:2111.04683. — [arXiv:2111.04683](https://arxiv.org/abs/2111.04683) find
  11. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Zhang, and 516 others. 202… — The Llama 3 Herd of Models
  12. Andreas Krause and Daniel Golovin. 2014. Submodular function maximization. In Lucas Bordeaux, Youssef Hamadi, and Kohli, editors, Tractability: Practical Approaches to Hard Problems, pages 71–104. Cambridge University Press, Cambridge. find
  13. Roger Grosse, Juhan Bae, Cem Anil, Nelson Elhage, Alex Tamkin, Amirhossein Tajdini, Benoit Steiner, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Kamilė Lukošiūtė, Karina Nguyen, Nicholas Joseph, Sam McCandlish, Jared Kaplan, and Samuel R. Bowman. 2023. Studying Large Language Model General… find
  14. Peijian Gu, Yaozong Shen, Lijie Wang, Quan Wang, Hua Wu, and Zhendong Mao. 2023. IAEval: A Comprehensive Evaluation of Instance Attribution on Natural Language Understanding. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 11966–11977, Singapore. Association for Compu… find
  15. Yuling Gu, Oyvind Tafjord, Bailey Kuehl, Dany Haddad, Jesse Dodge, and Hannaneh Hajishirzi. 2025. OLMES: A Standard for Language Model Evaluations. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 5005–5033, Albuquerque, New Mexico. Association for Computational Lingui… find
  16. Pang Wei Koh and Percy Liang. 2017. Understanding Black-box Predictions via Influence Functions. In Proceedings of the 34th International Conference on Machine Learning, pages 1885–1894. PMLR.
  17. Yongchan Kwon, Eric Wu, Kevin Wu, and James Zou. 2024. Datainf: Efficiently estimating data influence in lora-tuned llms and diffusion models. In The Twelfth International Conference on Learning Representations. find
  18. Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James Validad Miranda, Alisa Liu, Nouha Dziri, Xinxi Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Christopher Wilhelm, Luca Soldaini,… — Tulu 3: Pushing Frontiers in Open Language Model Post-Training
  19. Ikhtiyor Nematov, Dimitris Sacharidis, Katja Hose, and Tomer Sagi. 2024a. AIDE: Antithetical, Intent-based, and Diverse Example-Based Explanations. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 7(1):1051–1062. find
  20. Ikhtiyor Nematov, Dimitris Sacharidis, Tomer Sagi, and Katja Hose. 2024b. The susceptibility of example-based explainability methods to class outliers. CoRR, abs/2407.20678. — [arXiv:2407.20678](https://arxiv.org/abs/2407.20678) find
  21. Sung Min Park, Kristian Georgiev, Andrew Ilyas, Guillaume Leclerc, and Aleksander Madry. 2023. TRAK: attributing model behavior at scale. In International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, volume 202 of Proceedings of Machine Learning Research, pages… find
  22. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. "Why Should I Trust You?": Explaining the Predictions of Any Classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, pages 1135–1144, New York, NY, USA. Association for C…
  23. Jacob M. Schreiber, Jeffrey A. Bilmes, and William Stafford Noble. 2020. apricot: Submodular selection for data summarization in python. J. Mach. Learn. Res., 21:161:1–161:6. find
  24. Evan Pete Walsh, Luca Soldaini, Dirk Groeneveld, Kyle Lo, Shane Arora, Akshita Bhagia, Yuling Gu, Shengyi Huang, Matt Jordan, Nathan Lambert, Dustin Schwenk, Oyvind Tafjord, Taira Anderson, David Atkinson, Faeze Brahman, Christopher Clark, Pradeep Dasigi, Nouha Dziri, Allyson Ettinger, and 23 others… find
  25. Mengzhou Xia, Sadhika Malladi, Suchin Gururangan, Sanjeev Arora, and Danqi Chen. 2024. LESS: selecting influential data for targeted instruction tuning. In Forty-first International Conference on Machine Learning.
  26. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, and 22 others. 2024. Qwen2.5 technical report. CoRR, abs/2412.1511… find
  27. Chih-Kuan Yeh, Joon Kim, Ian En-Hsu Yen, and Pradeep K Ravikumar. 2018. Representer point selection for explaining deep neural networks. In Advances in Neural Information Processing Systems, volume 31. find
  28. Wei Zhang, Ziming Huang, Yada Zhu, Guangnan Ye, Xiaodong Cui, and Fan Zhang. 2021. On Sample Based Explanation Methods for NLP: Faithfulness, Efficiency and Semantic Evaluation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Join… find
  29. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models. CoRR, abs/2506.05176.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/schoenegger-2026-compact-example-based-explanations-language/