“It's a real, serious failing of human beings that we take 5 years to translate [Ascendance of a Bookworm,] a series of 33 books, while AI does it in an afternoon,” he said.

If I had to guess, most of those five years were spent on a handful of specially problematic chapters, while the bulk of the books was relatively straightforward.

Quof said he does not take MTPE jobs because “I think I have enough skill in Japanese and English that MTL tools do not currently improve my output. It would make my workflow harder because, as one used to providing 97% or higher accuracy, I would feel compelled to fix the AI's errors up to that high standard, and that would slow me down dramatically.” But over the years, Quof has had a habit of checking his own translations against AI tools.

My experience pretty much matches Quof's: it takes longer to fix a shitty machine translation than to do it by hand. It is however useful to check how ChatGPT or Google Translate would do it, specially for the problem bits.

I talked about it in the past. My work is involved with translating or working with translators so I can speak from first hand experience on this topic.We don't work with prose though. The quality of the translation tools and quality of the translated texts made such a huge jump in recent years that I don't really see it any worse than what a human translator gives us most of the time. It still needs human correction and QA by us but how much of it needs correction is getting less and less in huge leaps and bounds. And those tools are improving at a scary pace.

How well does that work for prose? I can't say. If we look at Tearoom Empire's change of translators in volume 9 (iirc) you can clearly see that there is an art to translating prose, and I wouldn't trust AI to translate at the highest level of translation the first Tearmoon translator produced. But it could easily see it producing a quality translation on the level of the second one. And if I think about all the light novels I've read so far, I would say that most of them are only mid at best in that regard. A computer can produce consistent quality where even the best human translator might just wing it here and there if they have a bad couple of weeks.

Then there is the problem with translators injecting their personal biases and political beliefs into the work. Just take the anime translation of Dragon Maid as an extreme example of what I mean.

example

I take a mid AI translation every time over an ideology injected version.

It makes sense that once someone has achieved fluency in both languages translation would be quite fast. I wonder at that point how much of a limiting factor typing speed is. Or perhaps the real time sink is just trying to figure out how to best phrase things?

I mean, it couldn't really hurt for consumers. until localizers stop "curating" the material to their personal tastes I don't see a reason to not bypass them.