Google Translate turns 20: What 2 decades of machine translation actually looks like

Google Translate turns 20: What 2 decades of machine translation actually looks like

10 0 0

Google Translate turned 20 this week. Twenty years of translating “hello” into 250 languages, mangling idioms, and occasionally saving your bacon during a trip abroad. The team put out a blog post with 20 fun facts to mark the occasion, and I went through it so you don’t have to.

It started in 2006 as a pure AI experiment. Not the flashy generative AI kind — the old-school statistical machine translation kind, where the system basically memorized patterns from bilingual documents and hoped for the best. It supported only a handful of languages back then. Arabic and Russian didn’t even show up until 2009. Today it covers almost 250 languages, which is genuinely impressive. For context, that’s more languages than most humans have heard of.

The service now handles over 100 billion words per day. That number is so large it stops meaning anything, so let me reframe it: that’s roughly 20 times the entire English Wikipedia, every single day. And it’s not just text — Translate now works across images, real-time voice, and even ambient conversation mode on Android. The camera translation feature alone processes over a billion images per month.

One thing that surprised me: Google claims Translate now preserves gender in translations for 90% of supported languages. I tested this with a few languages I know and it mostly held up, though it still defaults to masculine in ambiguous cases. Better than it was, but not solved.

They also highlighted that Translate added 110 new languages in 2024 alone, mostly through their Zero-Shot Machine Translation model. That’s the one that learns a new language without ever seeing parallel text for it. It works by training on related languages and then generalizing. Results vary wildly. For high-resource languages like Spanish or Japanese, it’s near flawless. For something like Bambara or Oromo, it gets the gist but the grammar can be a mess.

A few practical tips from the post that actually matter:

  • You can now tap and hold any translated text to hear it spoken with proper intonation, not the robotic monotone of five years ago.
  • The Transcribe mode on Android works offline for 60 languages now. This is huge if you travel.
  • The Gboard integration lets you translate mid-conversation without switching apps.

But let’s be honest: Google Translate still has problems. It’s terrible with context-dependent words (try translating “bank” into French and see if you get “river bank” or “financial bank”). It overcorrects for politeness in languages like Korean and Japanese, making you sound weirdly formal. And it still can’t handle poetry, sarcasm, or anything culturally specific without producing nonsense.

Still, for a free service that started as a side project in 2006, it’s remarkable. Twenty years ago, machine translation was a punchline. Now it’s infrastructure. I just wish they’d fix the damn idiom translations.

Comments (0)

Be the first to comment!