When to Ask, When to Answer: Strategic Clarification for Ambiguous Reference Resolution in Dialogue

Computer Science > Computation and Language arXiv:2611.05517 (cs) [Submitted on 3 Nov 2026] Title: When to Ask, When to Answer: Strategic Clarification for Ambiguous Reference Resolution in Dialogue Authors: Ibrahim Khebour, Nikhil Krishnaswamy.

计算机科学 > 计算与语言 arXiv:2611.05517 (cs) [提交于 2026 年 11 月 3 日] 标题：何时提问，何时回答：对话中歧义指代消解的策略性澄清 作者：Ibrahim Khebour, Nikhil Krishnaswamy。

Abstract: For humans, referential disambiguation is a collaborative process involving clarification questions asked when they will meaningfully change the outcome. However, large language models (LLMs) have been shown to be biased toward premature answering, preferring definite single-turn responses over more clarification exchanges.

摘要：对于人类而言，指代消解是一个协作过程，涉及在能够有意义地改变结果时提出的澄清问题。然而，研究表明大型语言模型（LLMs）倾向于过早回答，比起更多的澄清交流，它们更偏好确定的单轮响应。

In this paper, we study the problem of “strategic clarification” — asking clarification questions in a reference game when their answers will narrow the set of candidate referents. We construct a controlled test environment from the SIMMC 2.1 dataset that records how many questions each ambiguous reference needs to be successfully resolved, and examine how baseline LLM behavior and common post-training methods ask clarifying questions vs. attempt to answer the referent.

在本文中，我们研究了“策略性澄清”问题——即在指代游戏中，当问题的答案能够缩小候选指代对象集合时提出澄清问题。我们利用 SIMMC 2.1 数据集构建了一个受控测试环境，记录了每个歧义指代成功消解所需的问题数量，并考察了基准 LLM 行为及常见的训练后方法在提出澄清问题与尝试回答指代对象之间的表现差异。

We also propose BACE (Belief-Aware Critic with Epistemic tracking), a lightweight action chooser method on top of a frozen LM backbone, and show that BACE more consistently chooses the optimal number of clarifying questions than methods such as GRPO and ArCHer.

我们还提出了 BACE（基于信念感知的认知追踪评论家），这是一种在冻结的语言模型主干之上运行的轻量级动作选择方法，并证明了 BACE 比 GRPO 和 ArCHer 等方法能更一致地选择最优的澄清问题数量。