The parent comment is essentially correct that translations of the same material into different languages represent different views of the same data. A human translator must put in quite a bit of effort establishing what underlying situation is being described by a stretch of language.
Machine translations don't do this; they attempt to map one piece of language to another piece of (a different) language directly.
Relatedly, I tend to think of translations somewhat similar to a lossy system like those used in (say) image compression.
ie a compressed jpg of an image can retain quite a lot of the detail of the original, but it can introduce its own artifacts and lose some of the details
For things where the overall shape and picture is all that's required, that's fine. For things where the fine details matter, it's less fine.
This is a tempting illusion, but the evidence implies it’s false. Translation is simulation, not emulation.