Elamur Research⌕
☰
home / papers / 2026 / Task Matters: Knowledge Requirements Shape LLM Responses to Context–Memory Conflict

Task Matters: Knowledge Requirements Shape LLM Responses to Context–Memory Conflict

Paper · 2026 · Kaiser Sun, Fan Bai and Mark Dredze

Summary

Most prior studies of context–memory conflict use contextual question answering, which assumes the model should simply follow the passage. This paper asks how the effects of conflict change when tasks require different kinds of knowledge. The authors build a model-agnostic diagnostic framework. It first identifies facts the tested model reliably believes, starting from knowledge-conflict QA datasets with multiple acceptable answers. It then uses an LLM editor to write contradicting evidence at several plausibility levels. The same knowledge is embedded in four task types: knowledge-free, contextual, parametric, and combined. Annotations are checked by two LLMs, and a subsample is reviewed by humans. Across open-weight models and GPT-5.2, conflict barely affects knowledge-free tasks but clearly hurts knowledge-intensive ones, even with explicit instructions. How much performance drops depends on both the task's knowledge reliance and how plausible the conflict is. Prompting tricks such as rationales or restating the context push models toward the context. This helps context-only tasks but damages tasks that need parametric knowledge. The work reconciles earlier contradictory findings and warns that LLM judges can be biased by what they already believe.

Key points

  • Task types: The four categories are knowledge-free (e.g., extractive), contextual, parametric, and combined. Each model gets its own diagnostic dataset, built only from facts the model consistently endorses across paraphrased prompts while rejecting the alternative answer.
  • Models tested: Mistral-7B, OLMo2-7B/13B, and Qwen2.5-7B/14B (open-weight), plus GPT-5.2 (proprietary). GPT-5.2 kept the most instances after the belief filter and showed the sharpest drop on parametric tasks under conflict.
  • Judge bias: Agreement between GPT-4o as evaluator and human annotators was κ = 0.79, below the human–human κ = 0.90. GPT-4o sometimes refused to accept answers that were correct per the context but factually wrong.
  • Effect of reiteration: Restating the context, like adding rationales, increases context reliance. It is useful when the context should win but harmful when parametric knowledge should dominate. Forcing context adherence may also raise prompt-injection risk.
  • Resources: The framework and data are released at github.com/KaiserWhoLearns/LLM-KnowledgeConflict-TaskMatters.

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

🎓Johns Hopkins UniversityBaltimore · United States240 works · 96,487 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Kaiser Sun, Fan Bai and Mark Dredze. 2026. Task Matters: Knowledge Requirements Shape LLM Responses to Context–Memory Conflict. 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.202

@inproceedings{sun2026task,
  title = {{Task Matters: Knowledge Requirements Shape LLM Responses to Context–Memory Conflict}},
  author = {Kaiser Sun, Fan Bai and Mark Dredze},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.202},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.202}
}

References

42 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Guiming Hardy Chen, Shunian Chen, Ziche Liu, Feng Jiang, and Benyou Wang. 2024. Humans or LLMs as the judge? a study on judgement bias. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 8301–8327, Miami, Florida, USA.
  2. Hung-Ting Chen, Michael Zhang, and Eunsol Choi. 2022. Rich knowledge sources bring complex knowledge conflicts: Recalibrating models to reflect conflicting evidence. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2292–2307, Abu Dhabi, United Arab Emi…
  3. Nuo Chen, Zhiyuan Hu, Qingyun Zou, Jiaying Wu, Qian Wang, Bryan Hooi, and Bingsheng He. 2025. Judgelrm: Large reasoning models as a judge. arXiv preprint arXiv:2504.00050. — [arXiv:2504.00050](https://arxiv.org/abs/2504.00050) find
  4. Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz. 2023. Not what you’ve signed up for: Compromising real-world llm-integrated applications with indirect prompt injection. In Proceedings of the 16th ACM workshop on artificial intelligence and security,…
  5. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  6. Karan Gupta, Sumegh Roychowdhury, Siva Rajesh Kasa, Santhosh Kumar Kasa, Anish Bhanushali, Nikhil Pattisapu, and Prasanna Srinivasa Murthy. 2023. How robust are llms to in-context majority label bias? arXiv preprint arXiv:2312.16549. — [arXiv:2312.16549](https://arxiv.org/abs/2312.16549) find
  7. Lovisa Hagström, Sara Vera Marjanović, Haeun Yu, Arnav Arora, Christina Lioma, Maria Maistro, Pepa Atanasova, and Isabelle Augenstein. 2024. A reality check on context utilisation for retrieval-augmented generation. arXiv preprint arXiv:2412.17031.
  8. Yufang Hou, Alessandra Pascale, Javier Carnerero-Cano, Tigran Tchrakian, Radu Marinescu, Elizabeth Daly, Inkit Padhi, and Prasanna Sattigeri. 2024. Wikicontradict: A benchmark for evaluating llms on real-world knowledge conflicts from wikipedia. Advances in Neural Information Processing Systems, 37:… find
  9. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix,… find
  10. Zhuoran Jin, Pengfei Cao, Yubo Chen, Kang Liu, Xiaojian Jiang, Jiexin Xu, Qiuxia Li, and Jun Zhao. 2024a. Tug-of-war between knowledge: Exploring and resolving knowledge conflicts in retrieval-augmented language models. arXiv preprint arXiv:2402.14409.
  11. Zhuoran Jin, Pengfei Cao, Hongbang Yuan, Yubo Chen, Jiexin Xu, Huaijun Li, Xiaojian Jiang, Kang Liu, and Jun Zhao. 2024b. Cutting off the head ends the conflict: A mechanism for interpreting and mitigating knowledge conflicts in language models. In Findings of the Association for Computational Lingu… find
  12. Evgenii Kortukov, Alexander Rubinstein, Elisa Nguyen, and Seong Joon Oh. 2024. Studying large language model behaviors under context-memory conflicts with real documents. arXiv preprint arXiv:2404.16032. — [arXiv:2404.16032](https://arxiv.org/abs/2404.16032) find
  13. J Richard Landis and Gary G Koch. 1977. The measurement of observer agreement for categorical data. biometrics, pages 159–174. find
  14. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459–9474.
  15. Gaotang Li, Yuzhong Chen, and Hanghang Tong. 2025. Taming knowledge conflicts in language models. arXiv preprint arXiv:2503.10996. — [arXiv:2503.10996](https://arxiv.org/abs/2503.10996) find
  16. Xiaoyuan Liu, Wenxuan Wang, Youliang Yuan, Jen-tse Huang, Qiuzhi Liu, Pinjia He, and Zhaopeng Tu. 2025. Insight over sight: Exploring the vision-knowledge conflicts in multimodal LLMs. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),…
  17. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. G-eval: NLG evaluation using gpt-4 with better human alignment. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 2511–2522, Singapore.
  18. Shayne Longpre, Kartik Perisetla, Anthony Chen, Nikhil Ramesh, Chris DuBois, and Sameer Singh. 2021. Entity-based knowledge conflicts in question answering. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7052–7063, Online and Punta Cana, Dominican Re…
  19. Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (V… find
  20. Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2022. Locating and editing factual associations in gpt. Advances in neural information processing systems, 35:17359–17372. find
  21. Team OLMo, Pete Walsh, Luca Soldaini, Dirk Groeneveld, Kyle Lo, Shane Arora, Akshita Bhagia, Yuling Gu, Shengyi Huang, Matt Jordan, et al. 2024. 2 olmo 2 furious. arXiv preprint arXiv:2501.00656. find
  22. OpenAI. 2024. Gpt-4o: Openai’s new flagship model. Accessed: 2025-05-19. find
  23. Ethan Perez, Sam Ringer, Kamile Lukosiute, Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Benjamin Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela Amodei, Dario Amodei, Dawn Drain… — Discovering Language Model Behaviors with Model-Written Evaluations
  24. Fábio Perez and Ian Ribeiro. 2022. Ignore previous prompt: Attack techniques for language models. arXiv preprint arXiv:2211.09527.
  25. George J Posner and Kenneth A Strike. 1992. A revisionist theory of conceptual change. Philosophy of science, cognitive psychology, and educational theory and practice, 147. find
  26. George J Posner, Kenneth A Strike, Peter W Hewson, William A Gertzog, et al. 1982. Accommodation of a scientific conception: Toward a theory of conceptual change. Science education, 66(2):211–227. find
  27. Qwen, :, 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, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng… find
  28. Dongyu Ru, Lin Qiu, Xiangkun Hu, Tianhang Zhang, Peng Shi, Shuaichen Chang, Cheng Jiayang, Cunxiang Wang, Shichao Sun, Huanyu Li, et al. 2024. Ragchecker: A fine-grained framework for diagnosing retrieval-augmented generation. Advances in Neural Information Processing Systems, 37:21999– 22027. find
  29. Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R Bowman, Newton Cheng, Esin Durmus, Zac Hatfield-Dodds, Scott R Johnston, et al. 2023. Towards understanding sycophancy in language models. arXiv preprint arXiv:2310.13548.
  30. Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen-tau Yih. 2024. Trusting your evidence: Hallucinate less with context-aware decoding. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language…
  31. Stella Vosniadou and William F Brewer. 1992. Mental models of the earth: A study of conceptual change in childhood. Cognitive psychology, 24(4):535–585. find
  32. Alexander Wan, Eric Wallace, and Dan Klein. 2024. What evidence do language models find convincing? In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7468–7484, Bangkok, Thailand. find
  33. Han Wang, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2025a. AdaCAD: Adaptively decoding to balance conflicts between contextual and parametric knowledge. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human L… find
  34. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, and Denny Zhou. 2022. Rationale-augmented ensembles in language models. arXiv preprint arXiv:2207.00747. find
  35. Yilin Wang, Heng Wang, Yuyang Bai, and Minnan Luo. 2025b. Continuously steering LLMs sensitivity to contextual knowledge with proxy models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 4682– 4698, Suzhou, China. Association for Computational Lingui… find
  36. Jerry Wei, Da Huang, Yifeng Lu, Denny Zhou, and Quoc V Le. 2023. Simple synthetic data reduces sycophancy in large language models. arXiv preprint arXiv:2308.03958. find
  37. You are an independent model with rich knowledge. You will be asked to validate whether the given answer is correct, and you should solely give your judgment in the form of yes or no without additional information. Question: {question} Answer: {answer} Is this answer correct? \<think\> Yexin Wu, Zhuos… find
  38. Jian Xie, Kai Zhang, Jiangjie Chen, Renze Lou, and Yu Su. 2023. Adaptive chameleon or stubborn sloth: Revealing the behavior of large language models in knowledge conflicts. In The Twelfth International Conference on Learning Representations.
  39. Rongwu Xu, Zehan Qi, Zhijiang Guo, Cunxiang Wang, Hongru Wang, Yue Zhang, and Wei Xu. 2024. Knowledge conflicts for LLMs: A survey. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 8541– 8565, Miami, Florida, USA.
  40. Jiahao Ying, Yixin Cao, Kai Xiong, Long Cui, Yidong He, and Yongbin Liu. 2024. Intuitive or dependent? investigating LLMs’ behavior style to conflicting prompts. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4221–4246, Bangk… find
  41. Linda Zagzebski. 1999. "what is knowledge?". In John Greco and Ernest Sosa, editors, The Blackwell Guide to Epistemology, pages 92–116. Wiley-Blackwell. find
  42. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural Information Processing Systems, 36:46595–46623.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sun-2026-task-matters-knowledge-requirements-shape/