I’m mainly speaking from the viewpoint of translating (non basic dialog) from Japanese > English or Mandarin > English which often or not gets the results wrong or the translation is terrible (that is something you barely hear when you translate let’s say from German > English), you know the mistakes upon learning any language when you translate from the target one to your native tongue alongside the nuances, grammar and sentence structure.
Start with the difference between logographic language and alphabetic language. These are both written forms of spoken languages.
So part of the question will be: “is there a difference between spoken languages that first developed a logographic representation and spoken languages that first developed an alphabetic representation?”
In logographic languages, you’re not generally representing sounds, you’re representing concepts. In alphabetic languages, you’re generally not representing concepts, you’re representing sounds. If you invent a fundamentally new logograph, no one will know what it represents nor how to pronounce it until you tell them. If you develop a new letter, no one will know how to pronounce it until you tell them.
In an alphabetic language, you can combine letters to produce new words that may or may not have any relationship at all to similar words. The components in alphabetic languages essentially only carry a sound and no other meaning.
In a logographic language, you can combine logographs to produce new representations of concepts that generally have to relate to the concepts attached to the component logographs. Sometimes people can work out the new pronunciation as well.
Fundamentally, the ways in which we encode information in logographic versus alphabetic languages is wildly different. And since there are more concepts than could ever possibly be fully written down, these technologies need ways to expand. In alphabetic languages, we make new combinations of letter to make new words, but we also use idioms, phrases, and metaphors. Rarely do we make homophones and homonymns, but it happens. In logographic languages, nearly every logograph stands for multiple concepts by default.
At this point, it should start to become clear that if I am translating a text as text, without the context of the entire conversation, without hearing the spoken language, and without understanding the intent, subtext, poetic devices, and history of it all, then it’s very very easy to get totally lost when trying to translate.