In the other direction, I've seen gtl invert the meaning of sentences on subtle variations of the なければ pattern, which is probably a fairly small error in the problem space but, well, inverts the meaning of the sentence. A good reminder that parallel corpus machine learning does not see the world the same way you do.