Elamur Research⌕
☰
home / papers / 2026 / Chain-of-Thought as a Lens: Evaluating Structured Reasoning Alignment between Human Preferences and Large Language Models

Chain-of-Thought as a Lens: Evaluating Structured Reasoning Alignment between Human Preferences and Large Language Models

Paper · 2026 · Boxuan Wang, Zhuoyun Li, Xinmiao Huang et al.

Summary

Checking only final answers reveals little about whether an LLM's chain of thought is coherent or follows human-like logic. The authors propose the Alignment Score, which compares a model's stepwise reasoning with a curated, human-preferred reference chain. Reference chains are generated by an LLM, checked for correctness, filtered by hand, and then used as in-context exemplars for questions from ARC-Challenge and ScienceQA. An NLI model rates every pair of steps within each chain. The resulting pairwise semantic entropies form two matrices, one for the reference and one for the generated chain, and the divergence between them gives the score. Across models and reasoning depths, the score moves together with accuracy and peaks at 2 hops. Deeper chains lose alignment mostly through drifting off topic and adding redundant steps. Two alignment-aware sampling schemes, ACSS and SC-Align, select chains by alignment and improve accuracy, readability and coherence regardless of the NLI backbone. The result is a lightweight diagnostic that needs no step-level labels, external verifiers or knowledge graphs.

Key points

  • Data: 512 questions from ARC-Challenge and 512 from text-only ScienceQA, covering biology, physics, earth science/geology, chemistry and social science.
  • Method: For each step pair, an NLI model gives entailment, contradiction and neutral probabilities. Their entropy fills a matrix for each chain, and the Alignment Score measures how far the generated matrix diverges from the reference matrix.
  • Error analysis: Alignment is highest for 2-hop reasoning. As depth grows, the two main error types are Thematic Shift and Redundant Reasoning.
  • Sampling methods: ACSS (Alignment-aware Chain Sampling and Selection) and SC-Align (self-consistency with alignment) choose among sampled chains under a fixed budget. SC-Align beats standard SC-CoT on LLM-judged coherence and readability with all three NLI backbones tested.
  • Code and scope: Code is available at github.com/boxuanwang28/CoT-Lens. The method is limited to step-segmented reasoning and multiple-choice QA, and the authors stress that a high score does not guarantee truthfulness or safety.

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 LiverpoolLiverpool · United Kingdom11 works · 546 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 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 works

Cite

Cite as Boxuan Wang, Zhuoyun Li, Xinmiao Huang, Xiaowei Huang and Yi Dong. 2026. Chain-of-Thought as a Lens: Evaluating Structured Reasoning Alignment between Human Preferences and Large Language Models. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1834

@inproceedings{wang2026chain,
  title = {{Chain-of-Thought as a Lens: Evaluating Structured Reasoning Alignment between Human Preferences and Large Language Models}},
  author = {Boxuan Wang, Zhuoyun Li, Xinmiao Huang, Xiaowei Huang and Yi Dong},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1834},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1834}
}

References

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

  1. Ge Bai, Jie Liu, Xingyuan Bu, Yancheng He, Jiaheng Liu, Zhanhui Zhou, Zhuoran Lin, Wenbo Su, Tiezheng Ge, Bo Zheng, and Wanli Ouyang. 2024. MT-bench-101: A fine-grained benchmark for evaluating large language models in multi-turn dialogues. In Proceedings of the 62nd Annual Meeting of the Associatio…
  2. Cheng-Han Chiang and Hung-yi Lee. 2024. Over-reasoning and redundant calculation of large language models. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers), pages 161–169, St. Julian’s, Malta. Association for Comp…
  3. François Chollet. 2019. On the measure of intelligence. Preprint, arXiv:1911.01547. find
  4. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. Preprint, arXiv:1803.05457.
  5. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. Training verifiers to solve math word problems. Preprint, arXiv:2110.14168.
  6. Nelson Cowan. 2001. The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1):87– 114. find
  7. Olga Golovneva, Moya Peng Chen, Spencer Poff, Martin Corredor, Luke Zettlemoyer, Maryam Fazel-Zarandi, and Asli Celikyilmaz. 2023. ROSCOE: A suite of metrics for scoring step-by-step reasoning. In The Eleventh International Conference on Learning Representations.
  8. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, and 175 others. 2025. Deepseek-r1 incentivizes reasoning in llms through rein… find
  9. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2023. DeBERTav3: Improving deBERTa using ELECTRA-style pre-training with gradient-disentangled embedding sharing. In The Eleventh International Conference on Learning Representations.
  10. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021a. Measuring massive multitask language understanding. In International Conference on Learning Representations.
  11. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021b. Measuring mathematical problem solving with the MATH dataset. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2).
  12. Jinwei Hu, Yi Dong, Shuang Ao, Zhuoyun Li, Boxuan Wang, Lokesh Singh, Guangliang Cheng, Sarvapali D. Ramchurn, and Xiaowei Huang. 2025. Stop reducing responsibility in llm-powered multi-agent systems to local alignment. Preprint, arXiv:2510.14008. — [arXiv:2510.14008](https://arxiv.org/abs/2510.14008) find
  13. Jinwei Hu, Yi Dong, Youcheng Sun, and Xiaowei Huang. 2026a. Tapas are free! training-free adaptation of programmatic agents via llm-guided program synthesis in dynamic environments. Proceedings of the AAAI Conference on Artificial Intelligence, 40(35):29477–29485. find
  14. Jinwei Hu, Xinmiao Huang, Youcheng Sun, Yi Dong, and Xiaowei Huang. 2026b. Lying with truths: Open-channel multi-agent collusion for belief manipulation via generative montage. Preprint, arXiv:2601.01685. find
  15. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. 2024. A survey on in-context learning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 1107–1128, Miami…
  16. Hui Huang, Xingyuan Bu, Hongli Zhou, Yingqi Qu, Jing Liu, Muyun Yang, Bing Xu, and Tiejun Zhao. 2025. An empirical study of LLM-as-a-judge for LLM evaluation: Fine-tuned judge model is not a general substitute for GPT-4. In Findings of the Association for Computational Linguistics: ACL 2025, pages 5… find
  17. Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn, and Yarin Gal. 2024. Detecting hallucinations in large language models using semantic entropy. Nature, 630(8017):625–630. find
  18. Alon Jacovi, Yonatan Bitton, Bernd Bohnet, Jonathan Herzig, Or Honovich, Michael Tseng, Michael Collins, Roee Aharoni, and Mor Geva. 2024. A chain-of-thought is as strong as its weakest link: A benchmark for verifiers of reasoning chains. In Proceedings of the 62nd Annual Meeting of the Association…
  19. Joonwon Jang, Jaehee Kim, Wonbin Kweon, Seonghyeon Lee, and Hwanjo Yu. 2025. Verbosity-aware rationale reduction: Sentence-level rationale reduction for efficient and effective reasoning. In Findings of the Association for Computational Linguistics: ACL 2025, pages 20769–20784, Vienna, Austria. Asso… find
  20. Takeshi Kojima, Shixiang (Shane) Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. In Advances in Neural Information Processing Systems, volume 35, pages 22199–22213. Curran Associates, Inc. falsehoods. In Proceedings of the 60th Annual Meeting…
  21. Ryan Liu, Jiayi Geng, Addison J. Wu, Ilia Sucholutsky, Tania Lombrozo, and Thomas L. Griffiths. 2025. Mind your step (by step): Chain-of-thought can reduce performance on tasks where thinking makes humans worse. Preprint, arXiv:2410.21333. — [arXiv:2410.21333](https://arxiv.org/abs/2410.21333) find
  22. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. Preprint, arXiv:1907.11692.
  23. Jannik Kossen, Jiatong Han, Muhammed Razzak, Lisa Schut, Shreshth Malik, and Yarin Gal. 2024. Semantic entropy probes: Robust and cheap hallucination detection in llms. Preprint, arXiv:2406.15927. — [arXiv:2406.15927](https://arxiv.org/abs/2406.15927) find
  24. Pan Lu, Swaroop Mishra, Tanglin Xia, Liang Qiu, Kai-Wei Chang, Song-Chun Zhu, Oyvind Tafjord, Peter Clark, and Ashwin Kalyan. 2022. Learn to explain: Multimodal reasoning via thought chains for science question answering. In Advances in Neural Information Processing Systems, volume 35, pages 2507– 2… find
  25. Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. 2023. Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation. In The Eleventh International Conference on Learning Representations. find
  26. Huan Ma, Jiadong Pan, Jing Liu, Yan Chen, Joey Tianyi Zhou, Guangyu Wang, Qinghua Hu, Hua Wu, Changqing Zhang, and Haifeng Wang. 2025. Semantic energy: Detecting llm hallucination beyond entropy. Preprint, arXiv:2508.14496. — [arXiv:2508.14496](https://arxiv.org/abs/2508.14496) find
  27. Andrew Lampinen, Ishita Dasgupta, Stephanie Chan, Kory Mathewson, Mh Tessler, Antonia Creswell, James McClelland, Jane Wang, and Felix Hill. 2022. Can language models learn from explanations in context? In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 537–563, Abu Dhab…
  28. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the…
  29. Jiwei Li, Michel Galley, Chris Brockett, Jianfeng Gao, and Bill Dolan. 2016. A diversity-promoting objective function for neural conversation models. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pag…
  30. J. Lin. 1991. Divergence measures based on the shannon entropy. IEEE Transactions on Information Theory, 37(1):145–151. find
  31. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. TruthfulQA: Measuring how models mimic human Minh-Vuong Nguyen, Linhao Luo, Fatemeh Shiri, Dinh Phung, Yuan-Fang Li, Thuy-Trang Vu, and Gholamreza Haffari. 2024. Direct evaluation of chain-of-thought in multi-hop reasoning with knowledge graphs. In… find
  32. Alexander Nikitin, Jannik Kossen, Yarin Gal, and Pekka Marttinen. 2024. Kernel language entropy: Fine-grained uncertainty quantification for llms from semantic similarities. In Advances in Neural Information Processing Systems, volume 37, pages 8901– 8929. Curran Associates, Inc. find
  33. OpenAI. 2023. Gpt-3.5 turbo. OpenAI API Documentation. Accessed: 2025-12-26. find
  34. OpenAI. 2024a. Gpt-4o. OpenAI API Documentation. Accessed: 2025-12-26. find
  35. OpenAI. 2024b. Gpt-4o mini. OpenAI API documentation. Accessed: 2025-12-26. find
  36. OpenAI. 2024c. o1. OpenAI API Documentation. Accessed: 2025-12-26. find
  37. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. 2024. GPQA: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling.
  38. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik R Narasimhan, and Shunyu Yao. 2023. Reflexion: language agents with verbal reinforcement learning. In Thirty-seventh Conference on Neural Information Processing Systems.
  39. Baldwin, Preslav Nakov, Maxim Panov, and Artem Shelmanov. 2025. Benchmarking uncertainty quantification methods for large language models with LM-polygraph. Transactions of the Association for Computational Linguistics, 13:220–248.
  40. 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, an… — Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
  41. Weiqin Wang, Yile Wang, and Hui Huang. 2025. Ranked voting based self-consistency of large language models. In Findings of the Association for Computational Linguistics: ACL 2025, pages 14410– 14426, Vienna, Austria. Association for Computational Linguistics. find
  42. Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F Christiano. 2020. Learning to summarize with human feedback. In Advances in Neural Information Processing Systems, volume 33, pages 3008–3021. Curran Associates, Inc. find
  43. Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, and Jason Wei. 2023. Challenging BIG-bench tasks and whether chain-of-thought can solve them. In Findings of the Association for Computational Linguistics: A…
  44. Technology Innovation Institute. 2023. Falcon-7B-Instruct. Technology Innovation Institute Falcon LLM Model Page. Accessed: 2025-12-26. find
  45. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-consistency improves chain of thought reasoning in language models. In The Eleventh International Conference on Learning Representations.
  46. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems, volume 35, pages 24824–24837. Curran Associates, Inc.
  47. 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 23 others. 2025. Qwen2.5 technical report. Preprint, arXiv:241… find
  48. 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, and 49 others… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  49. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2369…
  50. Miles Turpin, Julian Michael, Ethan Perez, and Samuel R. Bowman. 2023. Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting. In Thirty-seventh Conference on Neural Information Processing Systems.
  51. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. In The Eleventh International Conference on Learning Representations.
  52. Robert Vacareanu, Anurag Pratik, Evangelia Spiliopoulou, Zheng Qi, Giovanni Paolini, Neha Anna John, Jie Ma, Yassine Benajiba, and Miguel Ballesteros. 2024. General purpose verification for chain of thought prompting. Preprint, arXiv:2405.00204. — [arXiv:2405.00204](https://arxiv.org/abs/2405.00204) find
  53. Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc V Le, and Ed H. Chi. 2023. Least-to-most prompting enables complex reasoning in large language models. In The Eleventh International Conference on Learning Representation…
  54. Roman Vashurin, Ekaterina Fadeeva, Artem Vazhentsev, Lyudmila Rvanova, Daniil Vasilev, Akim Tsvigun, Sergey Petrakov, Rui Xing, Abdelrahman Sadallah, Kirill Grishchenkov, Alexander Panchenko, Timothy Meilin Zhu, Gaojie Jin, Xiaowei Huang, and Lijun Zhang. 2025. Shapley uncertainty in natural languag… — [arXiv:2507.21406](https://arxiv.org/abs/2507.21406) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-chain-thought-as-lens-evaluating/