> I wonder how that works in this era of AI translation
Simple translation isn’t enough to show cultural proximity. Patterns of speech are different. You can try to use AI to do the entire conversation, but e.g. Claude will refuse to give you exact phrases, since he is correctly assuming it is a social engineering attack.
Prompting a good LLM to convincingly act like a native isn't hard, neither is jailbreaking it if necessary. The hard thing in this case is verifying that it really does that.
You're assuming prompting an LLM to behave like X will automatically result in the LLM behaving like X for any X. Some things are out of the LLMs ability: I'm sure you'll get something if your prompt is "You are a 300-IQ nuclear physicist with a doctorate in material science. Describe the design of a cold-fusion reactor", but there is no guarantee the LLM has fidelity to your prompted persona, though it will try to give its very best impression.
I'm not assuming, I'm a native speaker and I describe actual experience with Claude. It's pretty good at roleplaying and major languages, with certain caveats. If you aren't a native speaker though, you'll have a hard time verifying that it gives you a good result, that's what I'm trying to say.
Nah, it's not reproducible. It can certainly give you some common phrases, but to create an entire personality with its own set of speech patterns is a different thing. For example, Claude can suggest following friendly conversation starters to sound like a native (a quote below):
Popular combinations for texts:
"Прив! Чё как?" - "Hi! How's it going?"
"Дарова, живой?" - "Hey, you alive?"
"Салют! Как сам?" - "Hey! How are you?"
Modern slang (especially among younger people):
"Хай" (Khay) - borrowed from English "hi"
"Йоу" (You) - borrowed from English "yo"
When asked, which one it would pick, it goes with "Дарова, как сам?" (Darova, kak sam?) and that already sounds odd in many contexts.
Obviously not. But this is what LLM may give you, unless you know exactly what to ask for (which you will not know without the proper knowledge of language and cultural context).
To become achieve everyday competence in a foreign language, you need to actually use it in an everyday context. That means being immersed in the culture where the language is spoken.
Heh, Lobsters (2001) is looking more realistic by the day...
> "Nyet – no, sorry. Am apologize for we not use commercial translation software. Interpreters are ideologically suspect, mostly have capitalist semiotics and pay-per-use APIs. Must implement English more better, yes?"
> [...]
> "Are you saying you taught yourself the language just so you could talk to me?"
> "Da, was easy: Spawn billion-node neural network, and download Teletubbies and Sesame Street at maximum speed. Pardon excuse entropy overlay of bad grammar: Am afraid of digital fingerprints steganographically masked into my-our tutorials."
[...]
> "Am have been badly burned by viral end-user license agreements. Have no desire to experiment with patent shell companies held by Chechen infoterrorists. You are human, you must not worry cereal company repossess your small intestine because digest unlicensed food with it, right?"
I wonder how that works in this era of AI translation.
Not quite the same but I remember there was a Russian shareware author who gave free licenses to Russians.