Skip to content
conftrace_
2026 EMNLP EMNLP 2026

OVIBench: Benchmarking Online Video Question Answering under Interruption

AbstractRecent vision language models (VLMs) have achieved strong progress in video understanding. However, most existing video QA research and benchmarks still follow an offline, single-round paradigm, overlooking realistic interactions where users may interrupt the model during answer generation. To address this gap, we study online video QA under interruptions. We formulate the Online Video Question Answering under Interruption task and introduce OVIBench, the first standardized benchmark for this setting. OVIBench categorizes interruptions into three types: Cancellation, False Trigger, Correction and supports both open-ended and multiple-choice evaluations. To enable large-scale and reproducible testing, we develop an offline simulation protocol that reproduces interruption during generation under a unified temporal setup, together with a multi-dimensional metric suite for assessing interruption understanding and response generation. Experiments reveal clear weaknesses of mainstream VLMs in this setting, especially in reliably following user correction requests. Finally, we construct a train set OVI-Train for interruption-aware fine-tuning, which yields significant gains on OVIBench and validates the effectiveness of our benchmark and data design. OVIBench, OVI-Train, and the evaluation code will be released.

More from EMNLP 2026

All EMNLP 2026 papers →