A New View of Multi-modal Language Analysis: Audio and Video Features as Text “Styles”