Semantic Structure and Interpretability of Word Embeddings