Combating Word-level Adversarial Text with Robust Adversarial Training | IEEE Conference Publication | IEEE Xplore