Neural Interpretable Reasoning