Detecting (Un)answerability in Large Language Models with Linear Directions

2025-09-26

AI Summary

Large language models (LLMs) frequently generate hallucinated answers under information scarcity, necessitating reliable abstention detection mechanisms. This paper addresses the answerability classification task in extractive question answering. We propose an activation-based abstention detection method that identifies interpretable, linear directions in the hidden-layer activation space—learned via activation additions—which generalize across tasks and datasets. These directions are used to compute projection scores for binary answerability classification and further enable causal intervention analysis. Evaluated on two open-source LLMs (Llama-2 and Qwen) and four mainstream QA benchmarks, our method significantly outperforms existing prompt-engineering and supervised classifier approaches. It demonstrates strong robustness and generalization to diverse unanswerable cases, including factual gaps, ambiguous coreference, and logical contradictions.

Technology Category

• Natural Language Processing: Question Answering
• Machine Learning: Active Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Web Mining and Content Analysis: Community question answering

Abstract

Large language models (LLMs) often respond confidently to questions even when they lack the necessary information, leading to hallucinated answers. In this work, we study the problem of (un)answerability detection, focusing on extractive question answering (QA) where the model should determine if a passage contains sufficient information to answer a given question. We propose a simple approach for identifying a direction in the model's activation space that captures unanswerability and uses it for classification. This direction is selected by applying activation additions during inference and measuring their impact on the model's abstention behavior. We show that projecting hidden activations onto this direction yields a reliable score for (un)answerability classification. Experiments on two open-weight LLMs and four extractive QA benchmarks show that our method effectively detects unanswerable questions and generalizes better across datasets than existing prompt-based and classifier-based approaches. Moreover, the obtained directions extend beyond extractive QA to unanswerability that stems from factors, such as lack of scientific consensus and subjectivity. Last, causal interventions show that adding or ablating the directions effectively controls the abstention behavior of the model.

Problem

• Detecting unanswerable questions in large language models
• Identifying linear directions in activation space for classification
• Improving abstention behavior across diverse question types

Innovation

• Uses linear directions in activation space
• Applies activation additions during inference
• Projects hidden activations for classification score