The issue is if you're translating from a language that doesn't explicitly encode that information into a language that requires it; you need some way of fillin

The issue is if you're translating from a language that doesn't explicitly encode that information into a language that requires it; you need some way of filling in the blanks. This is all well and great when that information can be inferred from the explicit context provided; the real problem is that with translation there are a non-trivial number of cases that require the translator to consider the pragmatic context of what's being translated to figure out a correct/good translation (e.g. who's speaking/writing, who are they addressing, what is the overall societal context of content, what might be the purpose of the content, etc.). For a lot of these problems it's entirely reasonable to get a computer to fill in the blanks for 90%+ of cases, but the last few percent of cases require AI that is equivalent to that of a human.