GVLQA (Graph to Vision-Language Question Answering): A Vision-Language Graph Reasoning Dataset.