@steaphan

*takes a deep breath*

I am bilingual German/English. Born German, writing mostly in English. I sometimes translate my own stuff.

I have seen AI translations. Even human-edited AI translations.

They are - forgive my frankness - utter crap. An AI cannot capture the nuance, the connotations and the expressions of the target language.

It can only take the source language's words and turn them into target language words.

Working as editor with such a translation is a horror trip.

@Firlefanz @steaphan As an author and translator who has also wrestled with more than my share of this slop, trying to edit LLM-generated text is like an exercise in learned helplessness: I know I could do so much better if I translated from the ground up but I'm given neither the time nor money for that kind of effort, so I have to settle for "barely readable" with minimal edits and it almost physically hurts to let such a bad job pass through my hands.

LLM probability-translation can be a way to scan large amounts of foreign-language text, keeping the limitations of the technology in mind, but it doesn't do a good or even passable job of getting the fine points across accurately. Authors and readers deserve better, and so many works that could have been discovered and enjoyed by new audiences in translation will never get the chance now because they will have been reduced to rote reads bled of all originality and color, and often badly mistranslated as well.

@ljwrites @Firlefanz @steaphan

I, too, am an author-translator, and I don't bother with machine translation. For one thing, the texts I translate were usually written in 19th century German and use a lot of phrases that are simply no longer in use today.

For another, the translation itself is actually the _least_ part of my work - I must also explain all the local cultural, historical, and geographical contexts for these texts. And if I don't think about this during the translation process, the end result will be much worse.