As a non-native English speaker, let me tell you, terrible localization was very much a thing that happened well before machine translation, so that by itself (and more subtle typos or one-off errors) was definitely not enough to infer that someone had forgotten to fix a machine-translated line once.

You can definitely tell when something has been machine-translated and not fixed, but the real challenge is lack of context. This leads to nonsensical localization even today, whether it’s human or automated, especially in crowdsourced localizations, which are frequent in open source software. I contribute to some on occassion and maaaan, do I wish well intentioned people in that space would stop contributing to projects they don’t use/lines they haven’t seen in situ.