Uncovering Visual-Semantic Psycholinguistic Properties from the Distributional Structure of Text Embedding Space