Breaking down language barriers has long been a dream of science fiction — “Star Trek” had its Universal Translator to help the Enterprise crew understand exotic alien speech, and C3PO from “Star Wars” knew more than 6 million forms of communication from across the galaxy.
Now, thanks to advances in real-time translation software, the dream of speaking to anyone regardless of language is closer to reality than ever. Experts say human translators won’t be out of work anytime soon — they’re vital for legal proceedings, diplomatic discussions, and scenarios when exact word choice and tone are necessary — but new inexpensive digital tools allow people to speak easily in situations where communication once seemed impossible.
With software from the Austrian-based tech company iTranslate and a compatible set of wireless earphones, you can now have nearly 40 languages translated directly into your ear. But the tool doesn’t help users understand everything they’d hear on a crowded street yet. Currently, it’s focused on letting people speak with someone else using connected smartphones tethered to iTranslate-enabled earphones. It can facilitate basic transactions and everyday small talk between people who until recently couldn’t exchange a word.
“I think one area this is going to be really transformative in is business,” says Alexander Marktl, iTranslate’s CEO. “Business is also about trust, and if you know a person better because you talk about family or country and culture a little bit before, that can help to build trust.”
Real-time translation is a target for many big companies — Microsoft has built it into Skype video calls, Facebook can show posts in a user’s native language, and Google can translate websites or in-person conversations into dozens of languages. The creators of Travis the Translator, a handheld digital device that uses the best third-party translation software available for each of the 80 languages it supports, raised more than $730,000 during an IndieGogo campaign. They plan to ship devices in July.
As translation quality continues to improve, and such devices and apps become commonplace, it’ll be easier for travelers to share meaningful conversations and feel more confident walking into any restaurant in a foreign country. It will also be easier for people to understand the perspective of those from across the world, even without leaving home.
“From a social perspective, there is huge benefit to this technology,” says Ken Behan, vice president of sales at Systran, a translation company. “If you can read the newspaper in France and understand that point of view, that’s quite powerful for somebody who wants to learn more and get, a lot of times, a different opinion to what’s been reported locally in the U.S.”
While the quest to perfect translation technology stretches back decades — Systran was founded in 1968 based on Cold War-era machine translation research at Georgetown University — the recent explosion of virtual translators hinges on recent advances.
Initial computer translation efforts aimed to build complex rule sets and dictionaries that could translate every possible phrase or sentence. But recently, researchers have used libraries of translated materials and a wealth of multilingual information on the Internet, like newspapers and parliamentary documents from the European Union, to enable computers to figure out translations between two languages — almost like how a child learns to read by matching words with pictures.
“We go and find, let’s say, bilingual news articles that are written in one language and have been translated to another language, and are just out there on the web, or bilingual travel reviews — things like that,” says Mike Schuster, a senior staff research scientist at Google.
Similar algorithms that recently helped a computer best human world champions at the board game Go, and that taught self-driving cars the rules of the road, are now used to teach computers to understand how different languages match up. Known as neural networks, these algorithms are loosely modeled on the human brain and are designed to let computers learn to solve new problems without programmers having to spell out all the rules.
“For the last two or three years or so, they’ve just totally overrun everything as far as translation,” says Philipp Koehn, a computer science professor at Johns Hopkins University and author of "Statistical Machine Translation," a standard textbook in the field.
Software companies now let their powerful servers do the studying, and deliver real-time translations over the Internet in fractions of a second. As smartphones and other devices get more efficient, they’ll increasingly be able to do real-time translations without an online connection.
According to a paper published by Google researchers in September, the company’s neural net system generated 60 percent fewer errors than its previous method. Schuster said it’s been particularly useful for translating between English and East Asian languages, such as Korean and Japanese, and other languages that can confuse automated translators.
“For some of those languages, it seriously went from completely unusable to quite usable,” Schuster says.
“From a social perspective, there is huge benefit to this technology.”
Google has implemented the method for 42 of the 103 languages it translates and plans to roll it out for the rest this year. Schuster says the heightened accuracy has led to more people relying on its language apps and web tools — daily translations from Korean to English are up 75 percent over the past 6 months.
“This is not just because people use the phone more,” Schuster says. “This is because we launched this language pair there, and the quality has improved a lot, so it’s definitely useful.”
Schuster acknowledges that neither Google nor any other translation service is perfect. The latest systems make plenty of mistakes, including dropping parts of sentences or translating proper names as if they were ordinary words. Computer translation can’t compete with a bilingual human who understands the topic of a conversation, and scientists in the field say closing that gap is not a realistic goal for computers in the near future.
“It’s one thing teaching an AI engine to beat humans at Go, or chess, but translation is far more complex than that,” wrote Andy Way, a professor at Dublin City University and the editor of the journal Machine Translation, in an email. “Think how much world knowledge you bring to bear in processing a single sentence in your own language, and then at least double that when you want to translate.”
Human translators can also understand the tone and context of a conversation far better than any computer. And someone who isn’t bilingual may not be able to determine if the tone of a translation system’s output is appropriate, says David Rumsey, president of the American Translators Association, a trade group.
“You can have the same sentence that can be translated one way for one context or one customer, and then you can take the same sentence and translate it a different way for a different context or a different customer,” Rumsey says.
But the rapidly growing availability of better translation tools still stands to revolutionize how people communicate, at least as much as the social networking and messaging innovations of the past decade did, iTranslate’s Marktl says.
“It’s probably as transformative as something like Skype did or Facebook did,” he says. “I guess it will transform the world in a very big way.