Elamur Research⌕
☰
home / papers / 2026 / Structured Uncertainty guided Clarification for LLM Agents

Structured Uncertainty guided Clarification for LLM Agents

Paper · 2026 · Manan Suri, Puneet Mathur, Nedim Lipka et al.

Summary

Tool-calling LLM agents often fail when a user's request is vague or missing details. They either guess arguments or ask needless questions, because prompting-based clarification has no clear rule for what to ask or when to stop. The authors reframe clarification around the tool schema itself. The agent keeps probabilistic beliefs over candidate tool calls and their parameter domains, which separates ambiguity in the user's intent from the model's own predictive uncertainty. Each possible question is scored by Expected Value of Perfect Information (EVPI). A cost term based on aspects penalizes asking about things already covered.

This formulation drives two applications:

  • SAGE-Agent, which selects questions at inference time.
  • Uncertainty-weighted GRPO, which uses the uncertainty as a reward signal during training.

For evaluation they release ClarifyBench, a multi-turn benchmark with an LLM-based user simulator. SAGE-Agent covers ambiguous tasks better while asking fewer questions, and the training variant substantially lifts When2Call accuracy. This points to schema-grounded uncertainty as a useful basis for both more efficient agent interaction and more sample-efficient RL.

Key points

  • ClarifyBench has five domains (documents, vehicle control, stocks, travel, file system) and three query types (explicit, ambiguous, infeasible). It uses dynamic user simulation with follow-up requests, and part of its data comes from DocPilot logs.
  • Inference results: SAGE-Agent achieves 7–39% higher coverage on ambiguous tasks and asks 1.5–2.7× fewer questions than prompting and uncertainty-based baselines. It uses about 22K tokens and 54% fewer calls than Active Task Disambiguation.
  • Training results: uncertainty-weighted GRPO raises When2Call accuracy from 36.5% to 65.2% for Qwen2.5-3B and from 36.7% to 62.9% for Qwen2.5-7B. The trained 3B model beats the 7B model trained with standard GRPO (45.1%).
  • Redundancy penalty: raising λ from 0 to 0.5 cuts the number of questions by 18–27% with under 3% change in other metrics.
  • Other models: with Qwen2.5-14B-Instruct, SAGE-Agent reaches 54.56% ambiguous coverage versus 52.45% for ProCOT, and asks 1.41 questions versus 2.07.

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 Maryland, College ParkCollege Park · United States132 works · 25,380 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

Cite

Cite as Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi and Dinesh Manocha. 2026. Structured Uncertainty guided Clarification for LLM Agents. 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.2028

@inproceedings{suri2026structured,
  title = {{Structured Uncertainty guided Clarification for LLM Agents}},
  author = {Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi and Dinesh Manocha},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.2028},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.2028}
}

References

31 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jayachandu Bandlamudi, Ritwik Chaudhuri, Neelamadhav Gantayat, Kushal Mukherjee, Prerna Agarwal, Renuka Sindhgatta, and Sameep Mehta. 2025. A framework for testing and adapting rest apis as llm tools. find
  2. Maximillian Chen, Ruoxi Sun, Tomas Pfister, and Sercan Ö. Arık. 2025. Learning to clarify: Multi-turn conversations with action-based contrastive self-training. Preprint, arXiv:2406.00222. — [arXiv:2406.00222](https://arxiv.org/abs/2406.00222) find
  3. Michael Han Daniel Han and Unsloth team. 2023. Unsloth. find
  4. Yang Deng, Wenqiang Lei, Wenxuan Zhang, Wai Lam, and Tat-Seng Chua. 2022. Pacific: towards proactive conversational question answering over tabular and textual data in finance. arXiv preprint arXiv:2210.08817. find
  5. Yang Deng, Lizi Liao, Liang Chen, Hongru Wang, Wenqiang Lei, and Tat-Seng Chua. 2023. Prompting and evaluating large language models for proactive dialogues: Clarification, target-guided, and non-collaboration. Preprint, arXiv:2305.13626.
  6. Kung-Hsiang Huang, Akshara Prabhakar, Sidharth Dhawan, Yixin Mao, Huan Wang, Silvio Savarese, Caiming Xiong, Philippe Laban, and Chien-Sheng Wu. 2024a. Crmarena: Understanding the capacity of llm agents to perform professional crm tasks in realistic environments. arXiv preprint arXiv:2411.02305.
  7. Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. 2024b. Understanding the planning of llm agents: A survey. arXiv preprint arXiv:2402.02716. find
  8. Katarzyna Kobalczyk, Nicolas Astorga, Tennison Liu, and Mihaela van der Schaar. 2025. Active task disambiguation with llms. arXiv preprint arXiv:2502.04485. — [arXiv:2502.04485](https://arxiv.org/abs/2502.04485) find
  9. Sizhe Liu, Yizhou Lu, Siyu Chen, Xiyang Hu, Jieyu Zhao, Yingzhou Lu, and Yue Zhao. 2024. Drugagent: Automating ai-aided drug discovery programming through llm multi-agent collaboration. arXiv preprint arXiv:2411.15692. find
  10. Jiarui Lu, Thomas Holleis, Yizhe Zhang, Bernhard Aumayer, Feng Nan, Felix Bai, Shuang Ma, Shen Ma, Mengyu Li, Guoli Yin, and 1 others. 2024. Toolsandbox: A stateful, conversational, interactive evaluation benchmark for llm tool use capabilities. arXiv preprint arXiv:2408.04682.
  11. Chang Ma, Junlei Zhang, Zhihao Zhu, Cheng Yang, Yujiu Yang, Yaohui Jin, Zhenzhong Lan, Lingpeng Kong, and Junxian He. 2024. Agentboard: An analytical evaluation board of multi-turn llm agents. arXiv preprint arXiv:2401.13178. find
  12. Puneet Mathur, Alexa Siu, Varun Manjunatha, and Tong Sun. 2024. Docpilot: Copilot for automating pdf edit workflows in documents. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pages 232–246. find
  13. Changdae Oh, Seongheon Park, To Eun Kim, Jiatong Li, Wendi Li, Samuel Yeh, Xuefeng Du, Hamed Hassani, Paul Bogdan, Dawn Song, and Sharon Li. 2026. Uncertainty quantification in llm agents: Foundations, emerging challenges, and opportunities. Preprint, arXiv:2602.05073.
  14. Shishir G. Patil, Huanzhi Mao, Charlie Cheng-Jie Ji, Fanjia Yan, Vishnu Suresh, Ion Stoica, and Joseph E. Gonzalez. 2025. The berkeley function calling leaderboard (bfcl): From tool use to agentic evaluation of large language models. In Forty-second International Conference on Machine Learning. find
  15. Cheng Qian, Bingxiang He, Zhong Zhuang, Jia Deng, Yujia Qin, Xin Cong, Zhong Zhang, Jie Zhou, Yankai Lin, Zhiyuan Liu, and 1 others. 2024. Tell me more! towards implicit user intention understanding of language model driven agents. arXiv preprint arXiv:2402.09205. find
  16. Sudha Rao and Hal Daumé III. 2018. Learning to ask good questions: Ranking clarification questions using neural expected value of perfect information. arXiv preprint arXiv:1805.04655. find
  17. Hayley Ross, Ameya Sunil Mahabaleshwarkar, and Yoshi Suhara. 2025. When2Call: When (not) to call tools. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 3391– 3409…
  18. Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36:68539–68551.
  19. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. Preprint, arXiv:2402.03300.
  20. Sanidhya Vijayvargiya, Xuhui Zhou, Akhila Yerukola, Maarten Sap, and Graham Neubig. 2025. Interactive agents to overcome ambiguity in software engineering. find
  21. Chenyu Wang, Weixin Luo, Sixun Dong, Xiaohua Xuan, Zhengxin Li, Lin Ma, and Shenghua Gao. 2025. Mllm-tool: A multimodal large language model for tool agent learning. Preprint, arXiv:2401.10727. — [arXiv:2401.10727](https://arxiv.org/abs/2401.10727) find
  22. Wenxuan Wang, Juluan Shi, Chaozheng Wang, Cheryl Lee, Youliang Yuan, Jen-tse Huang, and Michael R Lyu. 2024. Learning to ask: When llms meet unclear instruction. arXiv preprint arXiv:2409.00557. — [arXiv:2409.00557](https://arxiv.org/abs/2409.00557) find
  23. Shirley Wu, Michel Galley, Baolin Peng, Hao Cheng, Gavin Li, Yao Dou, Weixin Cai, James Zou, Jure Leskovec, and Jianfeng Gao. 2025. Collabllm: From passive responders to active collaborators. Preprint, arXiv:2502.00640. find
  24. Jian Xie, Kai Zhang, Jiangjie Chen, Tinghui Zhu, Renze Lou, Yuandong Tian, Yanghua Xiao, and Yu Su. 2024. Travelplanner: A benchmark for real-world planning with language agents. arXiv preprint arXiv:2402.01622. find
  25. Jingjing Xu, Yuechen Wang, Duyu Tang, Nan Duan, Pengcheng Yang, Qi Zeng, Ming Zhou, and Xu Sun. 2019. Asking clarification questions in knowledge-based question answering. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confer… find
  26. Shunyu Yao, Noah Shinn, Pedram Razavi, and Karthik Narasimhan. 2024. tau-bench: A benchmark for tool-agent-user interaction in real-world domains. arXiv preprint arXiv:2406.12045.
  27. Guoli Yin, Haoping Bai, Shuang Ma, Feng Nan, Yanchao Sun, Zhaoyang Xu, Shen Ma, Jiarui Lu, Xiang Kong, Aonan Zhang, and 1 others. 2024. Mmau: A holistic benchmark of agent capabilities across diverse domains. arXiv preprint arXiv:2407.18961. find
  28. Yangyang Yu, Haohang Li, Zhi Chen, Yuechen Jiang, Yang Li, Denghui Zhang, Rong Liu, Jordan W Suchow, and Khaldoun Khashanah. 2024. Finmem: A performance-enhanced llm trading agent with layered memory and character design. In Proceedings of the AAAI Symposium Series, volume 3, pages 595–597. find
  29. Michael J. Q. Zhang, W. Bradley Knox, and Eunsol Choi. 2025. Modeling future conversation turns to teach llms to ask clarifying questions. Preprint, arXiv:2410.13788. find
  30. Michael JQ Zhang and Eunsol Choi. 2023. Clarify when necessary: Resolving ambiguity through interaction with lms. arXiv preprint arXiv:2311.09469. find
  31. Xuan Zhang, Yang Deng, Zifeng Ren, See Kiong Ng, and Tat-Seng Chua. 2024. Ask-before-plan: Proactive language agents for real-world planning. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 10836–10863. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/suri-2026-structured-uncertainty-guided-clarification-llm/