To be clear, I’m not talking about translating dialog such as “how are you” or “can you tell me where Barcelona is?” or anything equivalent of that instead, my main point is talking about translating the following types of dialog from English to Spanish involving:
- Profanity: “F*ck You!” used as a joke, rather than to offend
- Puns: As in terms that rhyme but consist of different words.
- Slang: I’m talking about colloqial speech or idioms
- Hyperbole: exaggerated dialog not meant to be literal
Like this for example: “I nearly died of embarrassment” but the translation put out “Casi me muero de vergüenza” which is just nonsense as it’s translated too literally (died > muero) when “died” is more on an idiomatic take on feeling embarrassed and emphasizes that the social blunder was incredibly awkward or uncomfortable. (Is the translation even right?)
Machine translation is not good at tone. So if a friendly fuck you is a different expression from one uttered in anger, I’d be very careful. Puns and slang only if it’s established enough. By which I mean is been around for a while and the model could train on it. Hyperbole will also be tough because if it’s devoid of tone how would the computer know? And even if you run instant spoken translation I wouldn’t trust any assistant to get it right.