How Good is Google Translate, Anyway?

This just in: Machine Learning is the Next Big Thing. It’s going to transform the way we work, the way we live, the way we do business. Basically– high caliber analysis here– machine learning will solve all our problems.

I’m kidding, of course. This isn’t news; machine learning’s life-changing (awe-inspiring, jaw-dropping, mind-blowing) potential has been the subject of anticipation and acclaim for some time. Google, for instance, announced its impact on most of the company’s core products and its hopes for future implementations of the technology as early as 2016.

Google “machine learning”, and the first page of news results alone lists a staggering array of uses. Machine learning is being used to locate the source of a Salmonella outbreak, develop new drugs, improve medical imaging, combat potholes, and find patterns in quantum fireworks.

One of the most vaunted applications of machine learning is machine translation. Historically, automated translation attempted to use encyclopedic collections of syntactic rules and vocabulary, with limited success. With machine learning, however, translation engines have improved drastically, with many predicting that effective automated translation could be achieved in mere years. Chief among these translation engines is Google Translate.

Yet as many a foreign language student has discovered, even Google Translate’s translations can be… iffy. Pass it anything more complicated than a few words, and an error is more likely than not. Google Translate’s strength is its ability to translate any word into (almost) any language, using its extensive database to map appropriate word pairings. But its grasp of context and connotation hasn’t, probably, replaced the need for dictionaries and thesauri in translation, as this article in the Atlantic last year illustrates.

This inherent inaccuracy has given rise to a whole new subgenre of comedy: making fun of Google Translate’s errors. The YouTube channel Translator Fails has over 800,000 subscribers, and The Tonight Show Starring Jimmy Fallon’s Google Translate Songs segments routinely rack up viewer counts in the millions.

Watch the video if you’re not convinced: whether in syntax or in substance, Google Translate’s translations ought not to be trusted.

Still, the tool’s accuracy is impressive considering the difficulty of the task at hand. What Google Translate has already accomplished would have been thought impossible at one point, and for some languages, its accuracy scores at over 90%. Is time the missing ingredient; will Google Translate in another ten or 20 years be what many have heralded already? Or is the engine failing to live up to expectations because those expectations are unfeasible; should we give up the ghost on machine translation?

For now, the answer might be c), none of the above. Machine translation tools can be incredibly valuable without replacing the role of a human translator. A start-up called Lilt is seeking to use machine translation tools, with machine learning at their core, to improve translator’s efficiency and effectiveness. The system provides suggestions for the next word or sentence; the human editor filters for quality, subtext, context, and style. This model combines the best of both worlds while assuring the accuracy that’s essential for translations in business, government, or publishing.

Maybe Machine Learning will solve the ages-old problem of automated translation– along with potholes– at some point, but for now, we need both man and machine.

7 thoughts on “How Good is Google Translate, Anyway?

  1. I agree, those who are making fun of the AI translators skill are definitely looking it from a “glass is half empty” (really only 1/4 empty) perspective. It’s amazing what these tools are capable of and they’re only getting better.


  2. Great post, loved the video! I had never heard of Lilt, but I love the idea. Maybe I’m biased but until a translator can consistently achieve above 99% accuracy I’m not going to put my trust in its output, especially when I can’t double check it. Having a human assist the machine takes this risk away. I also wonder if having a human edit the translator’s output is providing the machine learning algorithm with training data to improve. This would be a large amount of training data and maybe Lilt won’t need humans down the road!


  3. Interesting that some languages accuracy scores over 90%, but I still feel like technological translators, even with machine learning, will never be able to fully comprehend and translate everything from one language to another. Especially since languages are always changing and adapting. With all the different slang and expressions it would be difficult for technology to understand that something like “jump the gun” doesn’t have anything to do with a gun or jumping. I have no doubt that machine learning will help translations but I don’t think the accuracy will ever reach 100%.


  4. I find this post helpful and optimistic–humans cannot be replaced, at least with this. There’s something about in-person communication, whether that be the inflection, young lingo, or witty turn-of-phrase, that will always surpass machine learning translation. However, that is not to say automated translation is not incredibly helpful. Google Translate is the reason that I could communicate with individuals who spoke no English in a small town in Spain this summer. An example like that truly speaks to the value and success of such machine learning thus far


  5. Very interesting post. It will be awesome to see just how much more intelligent things like Google translate get over time. Over the summer, I saw an offline translator product at a store called B8ta. Basically, you hold a button and speak into it in one language, and it repeats what you said out loud in another language. This was the first time I had ever seen a product with such translating capabilities. Check it out at


  6. Machine Learning is such a hot topic right, but the majority of people still don’t really see why it works and how it will change the world. Google Translate can serve as a very concrete example. Even though it may never get to be as good as a human translator (but who knows?), it can definitely alleviate the workload of one, which is essentially the goal of ML- to free up humans from simple, daunting tasks for more innovations in the future.


  7. Cool post! I also find it interesting that Google Translate has these difficulties in translating larger blocks of text accurately. Though its existing capabilities are undeniably impressive, it is slightly surprising to me that it has not progressed even further. In my blog post for this week, I wrote about a Google speech-to-text system; that program could take context clues from the surrounding words to add correct capitalization, punctuation, spelling, etc. Though translating between languages with total accuracy is a tall order, I think it will be exciting to see how Google Translate continues to develop over the next few years.


Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s