Answering questions by reasoning over both text and image inputs simultaneously to find correct answers.