Incorporating Syntactic and Phonetic Information into Multimodal Word Embeddings Using Graph Convolutional Networks.