Google Translate turns 20: What 250 languages and two decades of AI look like

Google Translate turned 20 this week, and the company marked the occasion with a blog post full of fun facts and a few new features. I’ve been using this thing since it was a shaky statistical machine translation experiment back in 2006, so I have opinions.

Let’s start with the numbers. Google Translate now supports almost 250 languages. That’s up from the original handful of European languages it launched with. The timeline is genuinely impressive: from 2006 to 2016, it added about 60 languages. Then the neural machine translation switch happened, and suddenly it was adding languages at a pace that would make a polyglot dizzy.

What I find more interesting than raw language count is how they got there. The early days were all about parallel texts — the UN and European Parliament documents were gold mines. You’d feed the system millions of translated sentences and hope it learned patterns. It worked, barely. The results were comically bad. I remember translating “The spirit is willing but the flesh is weak” into Russian and back, getting something about vodka and meat.

Then came neural machine translation in 2016, and everything changed. Instead of statistical guesswork, the system actually learned to represent meaning. The quality jump was immediate and dramatic. Suddenly, Spanish to English translations were actually readable. Then French. Then German. Then the model learned to handle gendered languages better, and context actually mattered.

The 250-language milestone isn’t just about counting. It’s about what those languages are. Google has been adding languages that don’t have large digital footprints — languages spoken by smaller communities, indigenous languages, languages without standardized writing systems. They’re doing this with something called Zero-Shot Machine Translation, which lets the model translate languages it’s never explicitly seen paired examples of. That’s wild.

One thing that bothers me: the blog post glosses over the fact that translation quality varies enormously across language pairs. English to Spanish is great. English to Somali? Still rough. The model is only as good as the training data, and for languages with limited digital presence, that data is thin. Google’s been addressing this with community contributions and partnerships, but it’s not solved.

New features worth trying: The expanded Lens integration now handles more languages in real-time. Point your phone at a menu in Thai and it overlays English text. It’s not perfect — I’ve gotten some gloriously wrong translations of street food ingredients — but it’s good enough to avoid accidentally ordering durian when you wanted mango.

There’s also a new conversation mode that’s less awkward than before. The old version had this terrible back-and-forth latency that killed any natural flow. The new one is faster, though it still struggles with overlapping speech and strong accents. Don’t try it in a loud bar.

A personal pet peeve: Google Translate still can’t handle idioms worth a damn. “It’s raining cats and dogs” becomes a literal disaster in most languages. The model doesn’t understand that idioms are cultural artifacts, not words to be translated. This is a fundamental limitation of current approaches, and I don’t see it changing anytime soon.

The 20th anniversary also brought a redesign of the mobile app that’s actually decent. The interface is cleaner, the language switching is faster, and they finally put the history button somewhere I can find it without hunting. Small wins matter.

I’ve watched Google Translate go from a party trick to a genuinely useful tool. I’ve used it to navigate foreign hospitals, negotiate taxi fares, and read academic papers in languages I don’t speak. It’s not a replacement for learning a language — anyone who thinks so has never tried to have a real conversation through it — but it’s a remarkable crutch.

Twenty years is a long time in tech. Google Translate has outlasted Google Reader, Google+, and about forty other Google products I’ve forgotten. That’s because it actually solves a real problem, and it keeps getting better. Not perfect, but better.

Try the Lens feature on a foreign menu this week. You’ll either get dinner or a story.