Learning Effective and Interpretable Semantic Models using Non-Negative Sparse Embedding.