Answering the Unanswerable Is to Err Knowingly: Analyzing and Mitigating Abstention Failures in Large Reasoning Models
Yi Liu, Xiangyu Liu, Zequn Sun, Wei Hu
TL;DR
The paper addresses where large reasoning models err by abstaining on unanswerable questions. It reveals a misalignment between internal recognition of unanswerability and external responses, and introduces a two-stage method combining cognitive monitoring with inference-time intervention to improve abstention while preserving reasoning on solvable tasks. Through experiments on SUM and UMWP with diverse LRMs, the approach achieves higher abstention and more accurate justifications without compromising answer accuracy, and reduces token usage. The work advances trustworthy AI by reducing hallucinations and demonstrating that latent and behavioral signals of answerability can be leveraged to guide abstention in real time.
Abstract
Large reasoning models (LRMs) have shown remarkable progress on complex reasoning tasks. However, some questions posed to LRMs are inherently unanswerable, such as math problems lacking sufficient conditions. We find that LRMs continually fail to provide appropriate abstentions when confronted with these unanswerable questions. In this paper, we systematically analyze, investigate, and resolve this issue for trustworthy AI. We first conduct a detailed analysis of the distinct response behaviors of LRMs when facing unanswerable questions. Then, we show that LRMs possess sufficient cognitive capabilities to recognize the flaws in these questions. However, they fail to exhibit appropriate abstention behavior, revealing a misalignment between their internal cognition and external response. Finally, to resolve this issue, we propose a lightweight, two-stage method that combines cognitive monitoring with inference-time intervention. Experimental results demonstrate that our method significantly improves the abstention rate while maintaining the overall reasoning performance.