Just learned that AI translation models can handle over 200 languages now

I was poking around the latest updates from some open source translation projects and found out they have models that cover like 202 languages now. That blew my mind because I remember a few years ago even Google Translate only did around 100. I was reading this paper from the BigScience workshop where they trained a model called NLLB that actually works for stuff like Fijian and Lingala. My cousin is from Ghana and speaks Twi, and I thought there was no way an AI could handle that. But sure enough, I tested a sentence and it came out pretty close to what he would say. It got me thinking about how this could help small communities keep their languages alive. Has anyone else tried using those multilingual models for a less common language?

2 comments

2 Comments

evan_campbell16d ago

I mean yeah it's cool that they can do 202 languages now but let's not pretend a model trained on internet data really understands Fijian culture or how people actually talk.

jordan_anderson9016d ago

Fair point but tools can still help bridge gaps even if they're not perfect.