@meredith @hyrulian DeepL *always* prioritizes a coherent sentence over any level of accuracy. Maybe it's because I'm mostly familiar with a rather non-concordant language pair (English and Japanese), but DeepL is largely worthless for anything beyond the most simple textbook-level phrase construction. Google Translate produces "worse" (less coherent) translations between the two, but what it produces generally has a more logical path that I can follow, even if it also isn't what I would consider to be accurate.