Semantic Frames and Visual Scenes: Learning Semantic Role Inventories from Image and Video Descriptions

25859

Knowledge Triples