Interpreting Character Embeddings With Perceptual Representations: The Case of Shape, Sound, and Color