Elamur Research⌕
☰
home / papers / 2026 / ART: Adaptive Reasoning Trees for Explainable Claim Verification

ART: Adaptive Reasoning Trees for Explainable Claim Verification

Paper · 2026 · Sahil Wadhwa, Himanshu Kumar, Guanqun Yang et al.

Summary

LLMs can reach fact-checking verdicts, but their reasoning is opaque and hard to challenge. Chain-of-Thought rationales may not reflect how the model actually decided. The authors (Capital One / Stevens Institute of Technology) propose ART (Adaptive Reasoning Trees), which turns a claim into a tree of pro and con arguments that can be inspected and disputed.

A debater LLM, prompted as supporter or attacker, generates child arguments, and each argument receives an initial intrinsic strength. A separate judge LLM then compares every supporting argument against every attacking one. A Bradley–Terry model converts the outcomes of these pairwise contests into relative scores and uses them to recalibrate the intrinsic strengths. The strengths are aggregated from the leaves upward into a probability that the root claim is true.

The method was tested on MedClaim, StrategyQA and TruthfulQA with several open-source models. ART generally beat Direct Prompting, CoT and ArgLLM. It also produces a decision trace that users can audit and contest, which is valuable in high-stakes settings.

Key points

  • The pipeline has four stages:
    • argument generation;
    • pairwise ranking by an LLM-as-a-Judge;
    • Bradley–Terry strength calibration (λ = 0.5, versus λ = 0, which corresponds to ArgLLM);
    • bottom-up aggregation.
  • Tree depth is capped at two. Breadth helps more than depth: with D=1, B=2 and Llama 3.3 70B as evaluator, ART reached 80.2 on TruthfulQA (ArgLLM: 71.0) and 75.4 on StrategyQA (ArgLLM: 69.4). On MedClaim it reached 67.4, slightly above ArgLLM's 67.1.
  • The debater models tested were Llama 3.1 3B/8B Instruct, Mistral-7B Instruct-v0.2, Mixtral-8x7B Instruct-v0.1 and Llama 3.3 70B Instruct.
  • Compared with ArgLLM, the BT-calibrated scores push root probabilities further from 0.5, giving more decisive predictions. The paper reports this correcting mistaken rejections.
  • The main limitations are cost, since ART needs many LLM calls rather than one, and the risk of judge bias or prompt sensitivity. Proposed future work includes evidence grounding (e.g., Wikipedia) and learned calibration methods.

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

🎓Stevens Institute of TechnologyHoboken · United States10 works · 95 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Sahil Wadhwa, Himanshu Kumar, Guanqun Yang, Abbaas Alif Mohamed Nishar, Pranab Mohanty, Swapnil Shinde and Yue Wu. 2026. ART: Adaptive Reasoning Trees for Explainable Claim Verification. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.28

@inproceedings{wadhwa2026art,
  title = {{ART: Adaptive Reasoning Trees for Explainable Claim Verification}},
  author = {Sahil Wadhwa, Himanshu Kumar, Guanqun Yang, Abbaas Alif Mohamed Nishar, Pranab Mohanty, Swapnil Shinde and Yue Wu},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.28},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.28}
}

References

49 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Vaibhav Adlakha, Parishad BehnamGhader, Xing Han Lu, Nicholas Meade, and Siva Reddy. 2024a. Evaluating correctness and faithfulness of instruction-following models for question answering. Preprint, arXiv:2307.16877.
  2. Vaibhav Adlakha and 1 others. 2024b. Evaluating correctness and faithfulness of llm explanations. arXiv preprint arXiv:2401.04895. — [arXiv:2401.04895](https://arxiv.org/abs/2401.04895) find
  3. Isabelle Augenstein and 1 others. 2023b. Challenges in factuality for llms. In EMNLP. find
  4. Ben Carterette, Paul Bennett, Max Chickering, and Susan Dumais. 2008. Here or there: Preference judgments for relevance. In Proceedings of ECIR 2008. find
  5. Springer. ECIR 2018 Test of Time Honorable Mention. find
  6. Alvin Chan and 1 others. 2023. Chateval: Towards better llm-based evaluators through multi-agent debate. In ACL.
  7. Xinya Chen and 1 others. 2023. Benchmarking large language models in rag settings. arXiv preprint arXiv:2311.09383. — [arXiv:2311.09383](https://arxiv.org/abs/2311.09383) find
  8. Nan Du and 1 others. 2023. Improving factual accuracy via multi-agent debate. In NeurIPS. find
  9. Gabriel Freedman, Adam Dejl, Deniz Gorur, Xiang Yin, Antonio Rago, and Francesca Toni. 2025. Argumentative large language models for explainable and contestable claim verification. Preprint, arXiv:2405.02079. — [arXiv:2405.02079](https://arxiv.org/abs/2405.02079) find
  10. Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. 2021. Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies. Preprint, arXiv:2101.02235. find
  11. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 542 o… — The Llama 3 Herd of Models
  12. Nuno M. Guerreiro, Duarte Alves, Jonas Waldendorf, Barry Haddow, Alexandra Birch, Pierre Colombo, and André F. T. Martins. 2023. Hallucinations in large multilingual translation models. Preprint, arXiv:2303.16104.
  13. Zhijiang Guo, M. Schlichtkrull, and Andreas Vlachos. 2021. A survey on automated fact-checking. Transactions of the Association for Computational Linguistics, 10:178–206.
  14. Ayush Agrawal, Mirac Suzgun, Lester Mackey, and Adam Kalai. 2024. Do language models know when they’re hallucinating references? In Findings of the Association for Computational Linguistics: EACL 2024, pages 912–928, St. Julian’s, Malta. Association for Computational Linguistics. find
  15. Pepa Atanasova and 1 others. 2020. Generating fact-checking explanations. In ACL.
  16. Isabelle Augenstein, Timothy Baldwin, Meeyoung Cha, Tanmoy Chakraborty, Giovanni Luca Ciampaglia, David Corney, Renee DiResta, Emilio Ferrara, Scott Hale, Alon Halevy, Eduard Hovy, Heng Ji, Filippo Menczer, Ruben Miguez, Preslav Nakov, Dietram Scheufele, Shivam Sharma, and Giovanni Zagni. 2023a. Fac… — Factuality Challenges in the Era of Large Language Models
  17. Clément Henin and Daniel Le Métayer. 2022. Beyond explainability: justifiability and contestability of algorithmic decision systems. AI Soc., 37(4):1397–1410. find
  18. Christopher Hidey and 1 others. 2018. Unsupervised argument mining from online discussions. In ACL. find
  19. Jie Huang and 1 others. 2025a. Hallucinations in large language models: Taxonomy, evaluation, and mitigation. arXiv preprint arXiv:2501.00001. — [arXiv:2501.00001](https://arxiv.org/abs/2501.00001) find
  20. Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2025b. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems…
  21. Geoffrey Irving and 1 others. 2018. Ai safety via debate. arXiv preprint arXiv:1805.00899. find
  22. 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
  23. Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock,… find
  24. Di Jin, Eileen Pan, Nassim Oufattole, Wei-Hung Weng, Hanyi Fang, and Peter Szolovits. 2020. What disease does this patient have? a large-scale open domain question answering dataset from medical exams. Preprint, arXiv:2009.13081.
  25. Youngsoo Jung and 1 others. 2022. Critic llms: Self-improvement through critical feedback. arXiv preprint arXiv:2212.12345. — [arXiv:2212.12345](https://arxiv.org/abs/2212.12345) find
  26. Manuj Kant, Sareh Nabi, Manav Kant, Roland Scharrer, Megan Ma, and Marzieh Nabi. 2025. Towards robust legal reasoning: Harnessing logical llms in law. Preprint, arXiv:2502.17638. find
  27. Ahmad Khan and 1 others. 2024. Debate as persuasion: Improving llm truthfulness. arXiv preprint arXiv:2402.10011. — [arXiv:2402.10011](https://arxiv.org/abs/2402.10011) find
  28. Takeshi Kojima, Shixiang Gu, Alistair Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. arXiv preprint arXiv:2205.11916.
  29. Jooyoung Lee and 1 others. 2023. Prompt-and-rank for fact verification with llms. arXiv preprint arXiv:2305.12345. — [arXiv:2305.12345](https://arxiv.org/abs/2305.12345) find
  30. Francesco Leofante, Hamed Ayoobi, Adam Dejl, Gabriel Freedman, Deniz Gorur, Junqi Jiang, Guilherme Paulino-Passos, Antonio Rago, Anna Rapberger, Fabrizio Russo, Xiang Yin, Dekai Zhang, and Francesca Toni. 2024. Contestable ai needs computational argumentation. Preprint, arXiv:2405.10729. — [arXiv:2405.10729](https://arxiv.org/abs/2405.10729) find
  31. Patrick Lewis and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In NeurIPS.
  32. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. Truthfulqa: Measuring how models mimic human falsehoods. Preprint, arXiv:2109.07958.
  33. S. Maity and M. J. Saikia. 2025. Large Language Models in Healthcare and Medical Applications: A Review. Bioengineering (Basel), 12(6):631. find
  34. Julian Michael and 1 others. 2023. Debate helps: Multi-llm collaboration for evaluation. In ICLR. find
  35. Preslav Nakov and 1 others. 2021. Automated fact-checking for assisting human fact-checkers. Communications of the ACM. find
  36. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, and 262 others.… — GPT-4 Technical Report
  37. Melissa Kazemi Rad, Huy Nghiem, Andy Luo, Sahil Wadhwa, Mohammad Sorower, and Stephen Rawls. 2025. Refining input guardrails: Enhancing llm-as-ajudge efficiency through chain-of-thought fine-tuning and alignment. Preprint, arXiv:2501.13080.
  38. A. Rago, F. Toni, M. Aurisicchio, and P. Baroni. 2016. Discontinuity-free decision support with quantitative argumentation debates. In Proceedings of the Fifteenth International Conference on Principles of Knowledge Representation and Reasoning, KR’16, page 63–72. AAAI Press. find
  39. Bichen Shi and 1 others. 2016. Fact checking in community forums. In WWW. find
  40. Abhinav Singh and 1 others. 2024. Rethinking interpretability in the era of llms. arXiv preprint arXiv:2404.09999. — [arXiv:2404.09999](https://arxiv.org/abs/2404.09999) find
  41. James Thorne and 1 others. 2018. Fever: a large-scale dataset for fact extraction and verification. In NAACL.
  42. 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. Preprint, arXiv:2305.04388.
  43. Juraj Vladika, Ivana Hacajov’a, and Florian Matthes. 2025. Step-by-step fact verification system for medical claims with explainable reasoning. ArXiv, abs/2502.14765. find
  44. Sahil Wadhwa, Chengtian Xu, Haoming Chen, Aakash Mahalingam, Akankshya Kar, and Divya Chaudhary. 2024. Northeastern uni at multilingual counterspeech generation: Enhancing counter speech generation with llm alignment through direct preference optimization. Preprint, arXiv:2412.15453. find
  45. Xuezhi Wang and 1 others. 2022. Self-consistency improves chain-of-thought reasoning in llms. arXiv preprint arXiv:2203.11171.
  46. Yuntao Wang and 1 others. 2023. Synergy of multiple personas in llm collaboration. arXiv preprint arXiv:2311.06510. — [arXiv:2311.06510](https://arxiv.org/abs/2311.06510) find
  47. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In NeurIPS.
  48. Haodi Zhang, Jiahong Li, Yichi Wang, and Yuanfeng Song. 2023. Integrating automated knowledge extraction with large language models for explainable medical decision-making. In 2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 1710–1717. find
  49. Wayne Xin Zhao and 1 others. 2024. Explainability in large language models: A survey. arXiv preprint arXiv:2402.04644. — [arXiv:2402.04644](https://arxiv.org/abs/2402.04644)

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wadhwa-2026-art-adaptive-reasoning-trees-explainable/