How Good is Google Translate?

Google Translate now “handles” 103 languages, but how smart is it, really?

It’s part of human nature to laugh when other people make silly mistakes. In fact, some of the best jokes on the internet are the so-called “fails” we commit”:  backyard trampoline disasters and skateboarding fails come to mind.

These are the mistakes we make and accidents we suffer in the daily task of being human. Thanks to smartphones, much of it is captured on video and posted for the world to see.

Now there’s a new kind of fail, and this time the joke’s on Google. And the best part is, these fails should make us all feel smarter.

It’s About Time Something Came Along to Make Us All Feel Smart

That’s good, because the endless skateboard accidents, wedding bloopers, and backyard idiot-festivals are constantly reminding us that humans can be pretty stupid at times.

You Know What Happens Next

You Know What Happens Next

But this new type of fail should make us all feel quite brainy. That’s because it’s the type of fail that occurs when machines attempt human intelligence. The most visible example of this is Google Translate.

The “Fails” of Google Translate

What happens when Google Translate fails to correctly render the subtleties of language?  “Translation Fails“.

“Translation fails” are hilarious, which is why there are entire websites and youtube channels devoted to pointing them out. Plus, now that Google Translate is the go-to travel app for today’s connected adventurer, we have a whole new treasure trove of translation fails to amuse us.

Google Translate Might Not be Responsible for This, But it Might as Well Have

Google Translate Might Not be Responsible for This, But it Might as Well Have Been

The fails are all over the place, too. They range from the amusing, where pop stars become interchangeable…

“Lady Gaga” translates into Britney Spears”

…to the disconcerting:

“Your husband had a heart attack” becomes “your husband’s heart was attacked.”.

But perhaps worst of all, and most hilarious, song lyrics become unrecognizable gibberish. This is where the limitations of Google Translate really sink in. Thanks to one very inspired young woman who dedicates an entire Youtube channel to the lyric fails of Google Translate, we can take a closer look.

“Hail to Me”

Adele’s popular song “Hello” starts out with the simple phrase, “Hello, it’s me”.

Even with that seemingly straightforward phrase, Google Translate gets it wrong. The lyrics are translated from English into several other languages, then back into English. When run through the Google Translate mill in such a way, that first line becomes hilariously wrong:

“Hello, it’s me.”  —>  Hail to me”

Want the full song, as translated through a few layers of Google Translate? A few more precious gems from that song:

  • Google Translate: “But it’s Defense Secretary saying our two” (Original: And it’s no secret that the both of us…)
  • Google Translate: “Operating time” (original: are running out of time)
Hilarious Fails Brought to Your By Google Translate

Hilarious Fails Brought to Your By Google Translate

And, because we couldn’t resist, one from Queen’s “Bohemian Rhapsody”:

You know how the original goes:

“Is this the real life? Is this just fantasy? Caught in a landslide, no escape from reality.”

Google Translate turns all that into:

“What is life? Justice or bad? Avalanche traps, seto escapist.”

Gotta laugh!

Why Can’t Google Translate Handle Lyrics?

Let’s be fair. It’s easy to trick Google Translate by typing in lyrics, slang, or questions that might confuse even a human translator, taken out of context.

Google Translate is especially bad at translating lyrics, however, which points out its limitations in a serious way.


Lyrics are highly contextual, like life and language in general.

Lyrics are love songs. They’re sad. They’re joyous. They’re personal. Or they’re full of masked references. They may refer to historical events. They might mention current events.

In short, they represent our emotions: messy, unpredictable, personal, and very much prone to misinterpretation without a full understanding of context. And what is language, if not the expression of our feelings and thoughts?

The Problem With Context

The problem is, song lyrics are so contextual that translating them requires more expertise than even most human translators might offer. Lyrics, if they can be translated at all, require translators with a specialist’s knowledge of the audience, the lingo, the cultural context, and the underlying message the song artist is trying to convey in the first place.

Of course, it’s not just lyrics that require context for effective translation. These contextual cues are what makes all translation so very tricky: slang, idioms, and cultural references.

There’s also the fact that these special parts of the language are constantly in flux. Words and phrases come and go out of style faster than you can say “man bun”.


Like Language Trends, Fashion Changes Fast and Requires Context in Order to Make Sense

And so it goes with any translation job involving more than a quick travel-related question like “Where’s the train station?”. Context is everything.

But isn’t Google Translate supposed to learn about all that? Isn’t it based on artificial intelligence that picks up on all that?

Yes, Google Translate “Learns” Languages

Contrary to what a lot of people think, Google uses artificial intelligence to power its translation tool. In other words, it’s much more than a cloud-based series of language dictionaries and assembled rules. It should, in theory, be picking up on the type of contextual cues that make language so much richer than a mere string of words.

Here’s how it works, very much simplified of course:

Google Translate uses artificial intelligence (AI) to learn how humans use language. Feed it data and it slowly picks out linguistic patterns and “learns” languages. The idea is, the more data you feed in, the better the AI becomes.

Artificial Intelligence Eats Data to Learn How to be More Human

Artificial Intelligence Eats Data to Learn How to be More Human

The AI devours reams and reams of translated text every day, searching for patterns in our language. It’s similar to the way a toddler learns. Toddlers spend a few years receiving input until their brains finally make sense of it all by recognizing patterns.

To give their “baby” the input it requires in order to thrive, the engineers at Google first fed their translation machine lots of translated documents from the United Nations. After all, the UN has been in the “business” of translation for decades so there’s tons of material to work with. Plus, they have to be pretty good at it: global relations often hinge upon what goes on behind closed doors at the UN.

At first, Google Translate Learned from Translated UN Documents

At first, Google Translate Learned from Translated UN Documents

But now, with the advent of Big Data, Google Translate has better “food” for its AI baby. That should mean their Translate app is even better these days. However, as you’ve seen with its botched translation of Adele’s lyrics, that doesn’t yet seem to be the case!

Why Google Translate Can’t Handle the Tough Stuff Yet

Google translate gives you translations that are statistically likely. These translations are drawn from all the data “food” it’s been given: machine-driven observations of existing translations. Therefore, the answers it gives are only as good as the data it has consumed.

That means, until the AI machine has been fed enough data, it’s going to keep on making mistakes. The problem is, even the scientists at Google have no idea when their AI will finally have enough data to mimic the human brain. They’re not really even sure if it’s even possible!

That’s why, with Google Translate, there’s no claim to absolute correctness… just statistical likelihood.

So, when Google Translate encounters something highly contextual, idiomatic or trendy, it can only use what it knows to try and come up with a translation.

The Problem: Translation is More of an Art Than a Science

Google’s AI is using what’s called in linguistics “descriptive grammar” techniques.  Descriptive grammar simply looks at the way people are really communicating and tries to make order from those observations. It’s an approach to language that flies directly in the face of old-school “rules-based” theories of language, which see human language like a programming language, or math.

It’s much more than that, however.

Linguists are now discovering that language is just too messy, too fluid, and too contextual to be merely a set of rules. It’s more of an art than a science, in other words, which is why Google Translate is having such a hard time.

So, until Google amasses enough data to cover the equivalent of what’s stored in the typical human brain over a lifetime of nuanced and contextual communication, it’s going to keep on giving us hilarious fails.

Sure, the artificial intelligence that fuels Google Translate may someday catch up, but that’s not exactly a given. Amassing the incredible amount of data required to learn slang, idioms, and other nuances of language that aren’t rule-driven: that could take an eternity.

The point is: we simply don’t know how long this massive devouring of data will take before Google Translate gets it right.

Even Google Admits Google Translate Has Serious Limitations

Google is aware of its own limitations…

“Of course, for nuanced or mission-critical translations, nothing beats a human translator,”

~Franz Och, head of Google Translate

The complexity of language may always outpace the ability of machine algorithms to learn it. We simply don’t know yet. For now, consider this: it’s even hard for humans to keep up with the body of slang, newly fashionable ways of phrasing meaning, and cultural references (it’s why we have “Urban Dictionaries”, after all… so we can keep up).

Language is so rich with idioms and slang, even humans have a hard time keeping up.

Language is so rich with idioms and slang, even humans have a hard time keeping up.

Conclusion: Google Translation’s AI Still Has a Long Way to Go

The idea behind Google’s AI is, of course, that the algorithm will eventually learn all the idioms in the world, as well as the slang, the dialects, and the urban speech trends that erupt in every culture.

But, as thousands of amused app users have discovered, the algorithm isn’t quite there yet with the more complex nuances of language.

So for now, machine learning may pick up the basics of simplified, textbook speech where only a minimal degree of cultural context is expected, but delve deeper into life’s messiness and you start getting those hilariously inappropriate results.

But that’s OK – we could all use a laugh once in a while, especially if it’s at the expense of Google.

The Ultimate Checklist of Questions You Should Ask Every New Customer

Sign up to get your free copy of my New Customer Questionnaire.
It's a great resource for anyone who wants to get detailed info on branding, target audience, and goals from their new customers. I use it all the time in my line of work. 


The Ultimate Checklist of Questions You Should Ask Every New Customer


Sign up for my newsletter and get your free copy.

Thank you!