If you’ve ever used Google Translate or a similar service, then you’ll know that automatic translation can be a hit or a horrifying miss. It can end with either, “Oh, that’s actually right,” or “Hah! Come look at what it says this means!”
The crude translations of our current tools have long convinced us that we will always need humans to provide the most accurate results. But, as with most things nowadays, Artificial Intelligence (AI) is challenging what we think is possible to achieve with technology.
After decades of research (and many burned out engineers), we’ve finally reached the point where computers can translate with “human” accuracy, thanks to a little something called Machine Translation (MT). Here’s what you probably want to know about it.
Why is MT so hard to get right?
There are plenty of factors which make translating hard for both humans and machines. For starters, a lot of meanings are subjective. They can depend on the mood, context, and who they’re directed towards. Not to mention niche vocabulary, different ways to express the same thing, and the local idioms used in each country.
Here’s an example of simple translation confusion: One day a business leader emailed his remote employee to congratulate him by saying, “You’re killing it out there!” While we all know that’s a compliment, his Spanish worker took it literally and replied with a lengthy apology. If humans are confused by small exchanges like this, imagine what it’s like for machines.
Then there’s the question of translating in the context of the overall topic (in line with linguistic structures), instead of crudely mapping the equivalent of each sentence on its own. This stuff isn’t easy to teach an MT system, you know.
What are the benefits of MT anyway?
The first thing that pops into everyone’s mind is being able to get around in a foreign country. While easier travel is definitely an important benefit of MT, it’s merely scratching the surface of possibilities.
Imagine having a fluid conversation over a video call with your foreign coworker, where both of you are speaking your native language, and yet perfectly understanding each other. Imagine if international workers could seamlessly communicate with the locals, or if live news could be instantly translated into multiple languages. If you’re a business owner, imagine if your company chatbot could accommodate every user in their own language? Now that would definitely help you expand your market share.
So what’s been cooking in MT lately?
Companies from all over the globe have surfaced with borderline amazing inventions using neural networks and deep learning. From a small group of eclectic entrepreneurs releasing a handheld AI translator for travelers, to large companies offering businesses the chance to break language barriers using Neural Machine Translation (NMT).
Google Translate isn’t exactly sitting nearby quietly glaring at their competitors. They’ve also been using an NMT model to gradually increase the fluency and accuracy of their translations. But it’s time for Google to move over, because Microsoft recently made headlines with their own translator.
A team of Microsoft researchers announced their MT system can match human accuracy when translating news articles from Chinese to English. As some of you may know, Chinese is one of the hardest languages to learn (and to translate). Previous attempts have resulted in some pretty embarrassing translation blunders, like street signs reading “Beware of safety” and a restaurant menu item listed as “roasted husband”.
But this may no longer be a problem, thanks to Microsoft’s major milestone in one of the most challenging natural language processing tasks. Xuedong Huang, a technical lead in the team said in a blog post,
“Hitting human parity in a machine translation task is a dream that all of us have had. We just didn’t realize we’d be able to hit it so soon.”
Exciting stuff! Are you as pumped as we are about the future of AI? Join us and 1500+ experts in all things AI and voice technology. While you’re feeling generous, follow us on Twitter here for a steady stream of cool posts and event updates. (You’ll be glad you did. Promise.)