Trusting deep learning natural-language models via local and global explanations