MovieQA is a large-scale question-answering dataset about movies, which can be used to evaluate automatic story understanding capabilities for videos and text. The dataset contains 14,944 multiple-choice questions from 408 movies with high semantic diversity, and each question includes five options, only one of which is the correct answer.

Answers to these questions can be drawn from a variety of sources: video footage, subtitles, storyboards, subset scripts, and DVS.