Comment by yaky
2 days ago
I don't see LLMs being able to replace translators for less-spoken languages.
I know a translator between two Eastern European languages, and some jobs require use of specialized dictionaries. Using LLMs in such cases would be very unreliable and would require even more effort to check and correct than doing it correctly in the first place. Plus, I really doubt that US tech firms are training LLMs on language spoken by "only" 6 million people.
As for entertainment, anyone who grew up in Eastern Europe with pirated movies with nasal monotone translations, or machine-translated video games knows how much those take away from the experience. Sure, "AI could do better", but could it be consistent and capture cultural nuances and idioms, etc?
I use Google Translate quite a lot and it's pretty good even for obscure languages. The voice to text thing on youtube videos has improved a lot in the last five years.
Re. not training on obscure languages, the current thing seems to be to chuck all digital information available into the training so although they probably don't hire a human, the specialist dictionaries are probably in there.
Even more so for spoken-only languages.