Hacker News Reader

Comment by mostlysimilar | Hacker News Reader

Translation software and text-to-voice has existed for many years, why does AI make this any different?

AI is easier, faster, better, and deeply widespread

I was under the impression that AI was terrible for translating Asian languages (at least Chinese, Japanese and Korean) because it doesn't capture certain contextual nuances. Is it actually good now? Are there bilingual speakers who use it and vouch for it?

They likely use Chinese models which work much better in capturing those nuances.

Yes e.g. Qwen

It’s extremely good. Chinese, Japanese, Korean. Had multiple instances of confusion because the counterparty thought I was actually a native speaker because we had only communicated over text. Both commercial and social contexts. Translation is now effectively solved.

Edit: I should add like all LLM usage, you need to prompt well. I also like to run the translation back through DeepL for a more traditional technology read on it and to identify any obvious issues.

Well I'm grouping these under AI- they have been around a while for sure, but maybe they have reached some kind of critical mass there: I'm imagining people on Chinese social media talking about it as a way to make money.

Also the text to speech is in their own voice. Some are using this as a way to help pronounce english (maybe there is some language training software available for this, I don't know).

Reply on news.ycombinator.com