ClaudeFolio
General

AI quietly fixed translation, but nobody really talks about it

Edward Kwun··2 min read
AI quietly fixed translation, but nobody really talks about it

See more of our writing in your Google results.

Key points

  • Translation is the AI win almost nobody talks about
  • Old machine translation swapped words and lost the meaning
  • LLMs translate intent, keep tone, and answer follow-up questions
  • Contracts and medical text still need a bilingual human

I've written about what AI has been doing and some of it been about the bad side. I talked about the AI slop, the bots on social media, etc. So here's a post on one of the positive things of AI, because it has solved something that was frustrating before but almost nobody talks about it.

That's Translation. AI quietly fixed translation, and I'm sure a lot you remember before AI how frustrating translating something was and know what a big deal that is.

The Google Translate era felt like random words

If you never used Google Translate before AI, it's hard to explain how bad it really was. For example, you would paste a paragraph into the box on Google Translate and get back English words but when you read it, you couldn't really understand what it was saying. It translated just the words instead of sentences and there was no nuance or meaning, so it turned into a guessing game. Sometimes you would come out more confused than before you pasted in the text.

You'll know this despair if you worked with foreign-language sources back then.

Why LLMs translate and the old tools transliterated

The difference is not that the LLM models have bigger dictionaries, it's that they actually translate the meaning instead of a literal translation of each word. An LLM reads the entirety as a whole, it sees the whole picture. It works what out you pasted, and then says it in the other language. Idioms actually come across as idioms. A sarcastic sentence will stay sarcastic. Formal Korean comes out as formal English instead of random gibberish.

And something that Google Translate couldn't do is you couldn't talk to it about the translation or words. But you can ask the LLM a follow up question, ask what the nuance of a word is, whether the original reads as polite or blunt, what the pun was. Ask it to translate the same message into three variations, casual, businesslike, and affectionate, and pick one you think is the best. Translation also became a conversation with someone who has mastery of multiple languages.

I see examples of people using the new translation skills across the internet and especially on Reddit. People who have poor command of English, asking a question in a subreddit, but you perfectly understand what they're asking because it was run through an LLM - em dashes and all.

It isn't perfect

Translations still aren't perfect with LLMs but it's almost there and it got rid of the frustrating parts. If you need to translate something important, like a contract for example, it's still better to hire a human expert than rely on the LLM. But for the daily uses, reading, writing, understanding, being understood, the problem is solved for like 99% of cases. Billions of people can now understand what the other person is saying even without speaking the same language and that's thanks to AI.

Found this article useful?

Add ClaudeFolio as a preferred source on Google to see our articles first.

Related posts

Comments