Models that understand both visual content and natural language to describe or answer questions about video.