Low-resource NLP, adversarial robustness, interpretability, MT