Lune

ACL Findings2026

Structured Uncertainty guided Clarification for LLM Agents

Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha

Abstract

LLM agents with tool-calling capabilities often fail when user instructions are ambiguous or incomplete, leading to incorrect invocations and task failures. Existing approaches operate in unstructured language spaces, generating clarifying questions through prompting strategies that lack principled criteria for determining which questions to ask and when to stop. We introduce a principled formulation of structured uncertainty that operates directly over tool parameters and their domains, cleanly separating specification uncertainty (what the user wants) from model uncertainty (what the LLM predicts). Our formulation uses Expected Value of Perfect Information (EVPI) to quantify the disambiguation value of each potential question, balanced against aspect-based cost modeling that prevents redundant questioning. We demonstrate the versatility of this formulation through two applications. First, SAGE-Agent uses structured uncertainty for inference-time question selection, achieving 7-39% higher coverage on ambiguous tasks while reducing clarification questions by 1.5-2.7× compared to strong prompting and uncertaintybased baselines. Second, we show that structured uncertainty provides effective training signals: uncertainty-guided reward modeling boosts When2Call accuracy from 36.5% to 65.2% (3B model) and 36.7% to 62.9% (7B model) through uncertainty-weighted GRPO training, demonstrating more sample-efficient reinforcement learning for tool-calling agents. To enable evaluation, we present ClarifyBench, the first multi-turn dynamic tool-calling disambiguation benchmark. Our results establish structured uncertainty as a principled framework that improves both inference-time interaction efficiency and training-time sample efficiency in tool-augmented agents.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

4 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • TravelPlanner: A Benchmark for Real-World Planning with Language Agents cites Toolformer: Language Models Can Teach Themselves to Use Tools
  • Uncertainty-Aware Clarification in LLM Agents with Information Gain cites Ask-before-Plan: Proactive Language Agents for Real-World Planning
  • Uncertainty-Aware Clarification in LLM Agents with Information Gain cites Toolformer: Language Models Can Teach Themselves to Use Tools
  • Ask-before-Plan: Proactive Language Agents for Real-World Planning cites TravelPlanner: A Benchmark for Real-World Planning with Language Agents
  • Ask-before-Plan: Proactive Language Agents for Real-World Planning cites Prompting and Evaluating Large Language Models for Proactive Dialogues: Clarification, Target-guided, and Non-collaboration
  • Ask-before-Plan: Proactive Language Agents for Real-World Planning cites Toolformer: Language Models Can Teach Themselves to Use Tools
  • Ask-before-Plan: Proactive Language Agents for Real-World Planning cites PACIFIC: Towards Proactive Conversational Question Answering over Tabular and Textual Data in Finance
  • Prompting and Evaluating Large Language Models for Proactive Dialogues: Clarification, Target-guided, and Non-collaboration cites PACIFIC: Towards Proactive Conversational Question Answering over Tabular and Textual Data in Finance
  • Uncertainty Quantification in LLM Agents: Foundations, Emerging Challenges, and Opportunities cites TravelPlanner: A Benchmark for Real-World Planning with Language Agents
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers4

  1. Uncertainty Quantification in LLM Agents: Foundations, Emerging Challenges, and OpportunitiesChangdae Oh, Seongheon Park, To Eun Kim, Jiatong Li et al.ACL 2026 · 8 citations
  2. Asking What Matters: Reward-Driven Clarification for Software Engineering TasksSanidhya Vijayvargiya, Vijay Viswanathan, Graham NeubigICML 2026 · 3 citations
  3. Uncertainty-Aware Clarification in LLM Agents with Information GainMengyi DENG, Zhiwei Li, Xin Li, Tingyu ZHU et al.ICML 2026 · 1 citation
  4. AgentGym2: Benchmarking Large Language Model Agents in De-Idealized Real-World EnvironmentsZhiheng Xi, Dingwen Yang, Jiaqi Liu, Jixuan Huang et al.ACL 2026

Builds on11

  1. Toolformer: Language Models Can Teach Themselves to Use ToolsTimo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu et al.NeurIPS 2023 · 5,989 citations
  2. TravelPlanner: A Benchmark for Real-World Planning with Language AgentsJian Xie, Kai Zhang, Jiangjie Chen, Tinghui Zhu et al.ICML 2024 · 376 citations
  3. Prompting and Evaluating Large Language Models for Proactive Dialogues: Clarification, Target-guided, and Non-collaborationYang Deng, Lizi Liao, Liang Chen, Hongru Wang et al.EMNLP Findings 2023 · 55 citations
  4. PACIFIC: Towards Proactive Conversational Question Answering over Tabular and Textual Data in FinanceYang Deng, Wenqiang Lei, Wenxuan Zhang, Wai Lam et al.EMNLP 2022 · 25 citations
  5. Ask-before-Plan: Proactive Language Agents for Real-World PlanningXuan Zhang, Yang Deng, Zifeng Ren, See-Kiong Ng et al.EMNLP Findings 2024 · 14 citations

Related papers

  1. When and What to Ask: AskBench and Rubric-Guided RLVR for LLM ClarificationJiale Zhao, Ke Fang, Lu ChengACL Findings 2026 · 1 citation
  2. Ambig-SWE: Interactive Agents to Overcome Underspecificity in Software EngineeringSanidhya Vijayvargiya, Xuhui Zhou, Akhila Yerukola, Maarten Sap et al.ICLR 2026 · 35 citations
  3. TRACER: Trajectory Risk Aggregation for Critical Episodes in Agentic ReasoningSina Tayebati, Divake Kumar, Nastaran Darabi, Davide Ettori et al.ICML 2026 · 7 citations
  4. Active Task Disambiguation with LLMsKasia Kobalczyk, Nicolás Astorga, Tennison Liu, Mihaela van der SchaarICLR 2025
  5. Value of Information: A Framework for Human-Agent CommunicationYijiang River Dong, Tiancheng Hu, Zheng Hui, Caiqi Zhang et al.ACL 2026 · 8 citations
Dusk over the sea between two cliffs drawn in fine vertical lines