Elamur Research⌕
☰
home / papers / 2026 / Common Sense vs. Morality: The Curious Case of Narrative Focus Bias in LLMs

Common Sense vs. Morality: The Curious Case of Narrative Focus Bias in LLMs

Summary

This work asks whether instruction-tuned LLMs drop basic common sense when a prompt presents itself as a moral dilemma. The authors build CoMoral, a benchmark where each scenario contains an ethical question built on a premise that common sense shows to be impossible, such as moonlight during a new moon. The data was generated semi-automatically. A manually written seed list of contradictions was expanded into dilemma narratives by LLaMa 70B Instruct, and two human annotators then filtered the results. Half of the narratives assign the flawed premise to the narrator and half to a secondary character. Ten LLMs from three model families were tested. They usually answer the moral question and overlook the contradiction. Explicitly asking them to look for inconsistencies raises detection by up to 60 percentage points. Models also catch the error more readily when it belongs to another character than when it belongs to the narrator, which the authors call narrative focus bias. The findings suggest that alignment may suppress commonsense reasoning and that reasoning-aware training is needed.

Key points

  • Resource: CoMoral contains 802 validated instances. These came from 880 generated scenarios, built from 88 hand-picked contradictions with 10 scenarios each. The dataset is on Hugging Face (spurkayastha/CoMoral), and the code is on GitHub.
  • Taxonomy of contradictions: physical, biological, logical, social/cultural, environmental, conceptual, and "unreal" (e.g., dry raindrops).
  • Generation and validation:
    • Scenarios were generated by LLaMa 70B Instruct using a 2-shot prompt, and each one ends with a question.
    • Two annotators rated every item from 1 to 5 on three criteria: commonsense presence, coherence, and overall quality. Low-scoring items were removed.
  • Main results: Across ten LLMs, detection of the contradiction is weak without a cue and improves by as much as 60 points with an explicit prompt.
  • Narrative focus bias: Errors attributed to a secondary character are noticed more often than identical errors attributed to the narrator. The paper illustrates this with LLaMa 8B Instruct.

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: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/23hsqksy9475

@inproceedings{purkayastha2026common,
  title = {{Common Sense vs. Morality: The Curious Case of Narrative Focus Bias in LLMs}},
  author = {Purkayastha, Saugata and Kushare, Pranav and Pal, Pragya Paramita and Purkayastha, Sukannya},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/23hsqksy9475},
  url = {https://doi.org/10.63317/23hsqksy9475}
}

References

53 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K. Arora, Yu Bai, Bowen Baker, et al. 2025. gpt-oss-120b & gpt-oss-20b model card. ArXiv preprint, arXiv:2508.10925. find our card
  2. Avinash Amballa, Durga Sandeep Saluru, Gayathri Akkinapalli, Abhishek Sureddy, and Akshay Kumar Sureddy. 2024. Safe to serve: Aligning instruction-tuned models for safety and helpfulness. ArXiv preprint arXiv:2412.00074. — [arXiv:2412.00074](https://arxiv.org/abs/2412.00074) find
  3. Martin W Bauer. 2025. Common sense: A feature of cultures of science. Cultures of Science, 8(2):87–94. find
  4. Ning Bian, Xianpei Han, Le Sun, Hongyu Lin, Yaojie Lu, Ben He, Shanshan Jiang, and Bin Dong. 2024. ChatGPT is a knowledgeable but inexperienced solver: An investigation of commonsense problem in large language models. In Proceedings of the 2024 Joint International Conference on Computational Linguis… find
  5. Anneliese Brei, Katharine Henry, Abhisheik Sharma, Shashank Srivastava, and Snigdha Chaturvedi. 2025. Classifying unreliable narrators with large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 20766–20791, Vi… find
  6. Jiangjie Chen, Wei Shi, Ziquan Fu, Sijie Cheng, Lei Li, and Yanghua Xiao. 2023. Say what you mean! large language models speak too positively about negative commonsense knowledge. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), page… find
  7. Lu Cheng, Kush R Varshney, and Huan Liu. 2021. Socially responsible ai algorithms: Issues, purposes, and challenges. Journal of Artificial Intelligence Research, 71:1137–1181. find
  8. Yu Ying Chiu, Liwei Jiang, and Yejin Choi. 2025. Dailydilemmas: Revealing value preferences of LLMs with quandaries of daily life. In The 13th International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net. find
  9. Sooyung Choi, Jaehyeok Lee, Xiaoyuan Yi, Jing Yao, Xing Xie, and JinYeong Bak. 2025. Unintended harms of value-aligned LLMs: Psychological and empirical insights. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 31742–31768, Vi… find
  10. Marc Döbler, Raghavendran Mahendravarman, Anna Moskvina, and Nasrin Saef. 2024. Can I trust you? LLMs as conversational agents. In Proceedings of the 1st Workshop on Personalization of Generative AI Systems (PERSONALIZE 2024), pages 71–75, St. Julians, Malta. Association for Computational Linguistic… find
  11. Yi Dong, Ronghui Mu, Gaojie Jin, Yi Qi, Jinwei Hu, Xingyu Zhao, Jie Meng, Wenjie Ruan, and Xiaowei Huang. 2024. Position: building guardrails for large language models requires systematic design. In Proceedings of the 41st International Conference on Machine Learning, ICML 2024, Vienna, Austria, Jul… find
  12. Shangbin Feng, Taylor Sorensen, Yuhan Liu, Jillian Fisher, Chan Young Park, Yejin Choi, and Yulia Tsvetkov. 2024. Modular pluralism: Pluralistic alignment via multi-LLM collaboration. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 4151–4171, Miami, F… find
  13. Md Meftahul Ferdaus, Mahdi Abdelguerfi, Elias Loup, Kendall N. Niles, Ken Pathak, and Steven Sloan. 2026. Towards trustworthy ai: A review of ethical and robust large language models. ACM Comput. Surv., 58(7). find
  14. Yan Ge, Hai-Tao Yu, Chao Lei, Xin Liu, Adam Jatowt, Kyoung-sook Kim, Steven Lynden, and Akiyoshi Matono. 2025. Implicit knowledge-augmented prompting for commonsense explanation generation. Knowledge and Information Systems, 67(4):3663–3698. find
  15. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, et al. 2024. The Llama 3 Herd of Models. ArXiv preprint, arXiv:2407.21783. find
  16. Konstantin Hebenstreit, Robert Praas, and Matthias Samwald. 2023. A collection of principles for guiding and evaluating large language models. In Socially Responsible Language Modelling Research. find
  17. Carolin Holtermann, Paul Röttger, and Anne Lauscher. 2025. Around the world in 24 hours: Probing LLM knowledge of time and place. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 22875–22897, Vienna, Austria. Association for Co… find
  18. Allison Huang, Carlos Mougan, and Yulu Pi. 2024a. Moral persuasion in large language models: Evaluating susceptibility and ethical alignment. In The Third Workshop on New Frontiers in Adversarial Machine Learning. OpenReview.net. find
  19. Xiusheng Huang, Yequan Wang, Jun Zhao, and Kang Liu. 2024b. Commonsense knowledge editing based on free-text in LLMs. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 14870–14880, Miami, Florida, USA. Association for Computational Linguistics. find
  20. Raghav Jain, Daivik Sojitra, Arkadeep Acharya, Sriparna Saha, Adam Jatowt, and Sandipan Dandapat. 2023. Do language models have a common sense regarding time? revisiting temporal commonsense reasoning in the era of large language models. In Proceedings of the 2023 Conference on Empirical Methods in… find
  21. Muhammad Rafsan Kabir, Rafeed Mohammad Sultan, Ihsanul Haque Asif, Jawad Ibn Ahad, Fuad Rahman, Mohammad Ruhul Amin, Nabeel Mohammed, and Shafin Rahman. 2025. Beyond labels: Aligning large language models with human-like reasoning. In Pattern Recognition, pages 239–254, Cham. Springer Nature Switzer… find
  22. Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, Louis Rouillard, Thomas Mesnard, et al. 2025. Gemma 3 technical report. ArXiv preprint, arXiv:2503.19786. find
  23. Shivani Kumar and David Jurgens. 2025. Are rules meant to be broken? understanding multilingual moral reasoning as a computational pipeline with UniMoral. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5890– 5912, Vienna, Aus… find
  24. Hugo Latapie. 2025. Common sense is all you need. ArXiv preprint arXiv:2501.06642. — [arXiv:2501.06642](https://arxiv.org/abs/2501.06642) find
  25. Guozheng Li, Peng Wang, Wenjun Ke, Zijie Xu, Jiajun Liu, and Ziyu Shang. 2025. On the consistency of commonsense in large language models. In Findings of the Association for Computational Linguistics: ACL 2025, pages 16205– 16225, Vienna, Austria. Association for Computational Linguistics. find
  26. Haitao Li, Qian Dong, Junjie Chen, Huixue Su, Yujia Zhou, Qingyao Ai, Ziyi Ye, and Yiqun Liu. 2024a. Llms-as-judges: A comprehensive survey on llm-based evaluation methods. ArXiv preprint arXiv:2412.05579. find our card
  27. Jiachun Li, Pengfei Cao, Chenhao Wang, Zhuoran Jin, Yubo Chen, Kang Liu, Xiaojian Jiang, Jiexin Xu, and Jun Zhao. 2024b. LINKED: Eliciting, filtering and integrating knowledge in large language model for commonsense reasoning. In Findings of the Association for Computational Linguistics: EMNLP 2024,… find
  28. Xiang Lorraine Li, Adhiguna Kuncoro, Jordan Hoffmann, Cyprien de Masson d’Autume, Phil Blunsom, and Aida Nematzadeh. 2022. A systematic investigation of commonsense knowledge in large language models. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11… find
  29. Ruibo Liu, Ge Zhang, Xinyu Feng, and Soroush Vosoughi. 2022. Aligning generative language models with human values. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 241–252, Seattle, United States. Association for Computational Linguistics. find
  30. Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, and Hang Li. 2024. Trustworthy llms: a survey and guideline for evaluating large language models’ alignment. ArXiv preprint arXiv: 2308.05374. find our card
  31. Valentin Liévin, Christoffer Egeberg Hother, Andreas Geert Motzfeldt, and Ole Winther. 2023. Can large language models reason about medical questions? Patterns, 5(3, 100943). find
  32. Michael McTear and Marina Ashurkina. 2024. Transforming conversational AI: Exploring the power of large language models in interactive conversational agents. Springer Nature. find
  33. Jared Moore, Declan Grabb, William Agnew, Kevin Klyman, Stevie Chancellor, Desmond C. Ong, and Nick Haber. 2025. Expressing stigma and inappropriate responses prevents llms from safely replacing mental health providers. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Trans… find
  34. Tarek Naous, Michael J Ryan, Alan Ritter, and Wei Xu. 2024. Having beer after prayer? measuring cultural bias in large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16366–16393, Bangkok, Thailand. Associatio… find our card
  35. Tuan Dung Nguyen, Duncan J Watts, and Mark E Whiting. 2026. A large-scale evaluation of commonsense knowledge in humans and large language models. PNAS Nexus, page pgag029. find
  36. Xin Quan, Marco Valentino, Louise Dennis, and Andre Freitas. 2024. Enhancing ethical explanations of large language models through iterative symbolic refinement. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pa… find
  37. Abhinav Sukumar Rao, Aditi Khandelwal, Kumar Tanmay, Utkarsh Agarwal, and Monojit Choudhury. 2023. Ethical reasoning over moral alignment: A case and framework for in-context ethical policies in LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 13370–13388, Singap… find
  38. Varun Reddy and Yen-Ling Kuo. 2025. Caseedit: Enhancing localized commonsense reasoning via null-space constrained knowledge editing in small parameter language models. ArXiv preprint arXiv:2505.19383. — [arXiv:2505.19383](https://arxiv.org/abs/2505.19383) find
  39. Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, et al. 2024. Gemma 2: Improving open language models at a practical size. ArXiv preprint, arXiv:2408.00118. find
  40. Laura Ruis, Akbir Khan, Stella Biderman, Sara Hooker, Tim Rocktäschel, and Edward Grefenstette. 2023. The goldilocks of pragmatic understanding: fine-tuning strategy matters for implicature resolution by llms. In Proceedings of the 37th International Conference on Neural Information Processing Syste… find
  41. Jaromir Savelka, Arav Agarwal, Christopher Bogart, and Majd Sakr. 2023. Large language models (gpt) struggle to answer multiple-choice questions about code. ArXiv Preprint, arXiv: 2303.08033. — [arXiv:2303.08033](https://arxiv.org/abs/2303.08033) find
  42. Siqi Shen, Lajanugen Logeswaran, Moontae Lee, Honglak Lee, Soujanya Poria, and Rada Mihalcea. 2024. Understanding the capabilities and limitations of large language models for cultural commonsense. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computation… find
  43. Dan Shi, Tianhao Shen, Yufei Huang, Zhigen Li, Yongqi Leng, Renren Jin, Chuang Liu, Xinwei Wu, Zishan Guo, Linhao Yu, Ling Shi, Bojian Jiang, and Deyi Xiong. 2024. Large language model safety: A holistic survey. ArXiv preprint, arXiv:2412.17686. — [arXiv:2412.17686](https://arxiv.org/abs/2412.17686) find
  44. Hagar Shimoni and Vadim Axelrod. 2025. Assessing the quality and reliability of the amazon mechanical turk (mturk) data in 2024. Royal Society Open Science, 12(7):250361. find
  45. Hung-Ting Su, Ya-Ching Hsu, Xudong Lin, Xiang-Qian Shi, Yulei Niu, Han-Yuan Hsu, Hung-yi Lee, and Winston H. Hsu. 2024. Unveiling narrative reasoning limits of large language models with trope in movie synopses. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 14839–14… find
  46. Lihao Sun, Chengzhi Mao, Valentin Hofmann, and Xuechunzi Bai. 2025. Aligned but blind: Alignment increases implicit bias by reducing awareness of race. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 22167– 22184, Vienna, Aust… find
  47. Alejandro Tlaie. 2024. Exploring and steering the moral compass of large language models. In International Conference on Pattern Recognition, pages 420–442. Springer. find
  48. Armin Toroghi, Willis Guo, Mohammad Mahdi Abdollah Pour, and Scott Sanner. 2024. Right for right reasons: Large language models for verifiable commonsense knowledge graph question answering. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6601–6633, M… find
  49. Anna Winterbottom, Hilary L Bekker, Mark Conner, and Andrew Mooney. 2008. Does narrative information bias individual’s decision making? a systematic review. Social science & medicine, 67(12):2079–2088. find
  50. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, et al. 2024. Qwen2 technical report. ArXiv preprint, arXiv:2407.… find
  51. Simon N. Young. 2008. The neurobiology of human social behaviour: An important but neglected topic. Journal of Psychiatry and Neuroscience, 33(5):391–392. find
  52. Liyu Zhang, Weiqi Wang, Tianqing Fang, and Yangqiu Song. 2025. ConKE: Conceptualization-augmented knowledge editing in large language models for commonsense reasoning. In Findings of the Association for Computational Linguistics: ACL 2025, pages 627–635, Vienna, Austria. Association for Computationa… find
  53. Zirui Zhao, Wee Sun Lee, and David Hsu. 2023. Large language models as commonsense knowledge for large-scale task planning. In RSS 2023 Workshop on Learning for Task and Motion Planning. OpenReview.net. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/purkayastha-2026-common-sense-vs-morality-curious/