Can Video LLMs Refuse to Answer? Alignment for Answerability in Video Large Language Models

2025-07-07

AI Summary

Current video large language models (Video-LLMs) lack the capability to discriminate question answerability, frequently generating hallucinated responses to questions beyond the video content. To address this, we propose the **Answerability Alignment framework**, the first systematic formalization and modeling of answerability judgment in video question answering. Methodologically, we design a dedicated evaluation protocol and a data generation pipeline; formulate an end-to-end answerability alignment objective grounded in video–caption pairs; and integrate an explicit refusal mechanism. Experiments demonstrate substantial improvements in both answerability detection accuracy and refusal rate for unanswerable questions. Our approach achieves state-of-the-art performance across multiple benchmarks and exhibits strong cross-dataset generalization.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Natural Language Processing: Question Answering
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

In the broader context of deep learning, Multimodal Large Language Models have achieved significant breakthroughs by leveraging powerful Large Language Models as a backbone to align different modalities into the language space. A prime exemplification is the development of Video Large Language Models (Video-LLMs). While numerous advancements have been proposed to enhance the video understanding capabilities of these models, they are predominantly trained on questions generated directly from video content. However, in real-world scenarios, users often pose questions that extend beyond the informational scope of the video, highlighting the need for Video-LLMs to assess the relevance of the question. We demonstrate that even the best-performing Video-LLMs fail to reject unfit questions-not necessarily due to a lack of video understanding, but because they have not been trained to identify and refuse such questions. To address this limitation, we propose alignment for answerability, a framework that equips Video-LLMs with the ability to evaluate the relevance of a question based on the input video and appropriately decline to answer when the question exceeds the scope of the video, as well as an evaluation framework with a comprehensive set of metrics designed to measure model behavior before and after alignment. Furthermore, we present a pipeline for creating a dataset specifically tailored for alignment for answerability, leveraging existing video-description paired datasets.

Problem

• Video-LLMs cannot reject irrelevant questions effectively
• Lack training to assess question relevance to video content
• Need framework to evaluate and refuse unfit questions

Innovation

• Alignment framework for question relevance evaluation
• Dataset pipeline for answerability training
• Comprehensive metrics for model behavior assessment