VisualSem: a high-quality knowledge graph for vision and language