Revisiting Negation in Neural Machine Translation