Explaining the Deep Natural Language Processing by Mining Textual Interpretable Features