Say what? Machine-based translation still makes significant errors. But new techniques from Google, Microsoft, and Facebook are much better, as shown by a new Hostelworld chat tool based on the tech.
Six months ago, Google began providing an enhanced online translation service for foreign language texts that has leapfrogged in fluency, accuracy, and speed over its previous service.
Google Translate says its service is improving more this year than in the last ten years combined.
Travel businesses gain
Slowly but surely, online travel sellers will soon see the benefits, as more accurate translations are enabling clearer communication and more bookings.
A good example of a travel company using the new tech is Hostelworld, a hostel booking platform, which on Wednesday added a tool to its iOS and Android mobile app that relies on Google’s enhanced translation tool.
Users pick a language, talk into their smartphone’s microphone, and then wait for words to be translated audibly and into text. A silly promotional video makes the point:
Google is the first of many tech companies to take advantage of so-called “neural machine translation,” which takes a fresh approach to the task of machine-based translations. Here’s an oversimplified explanation: The computer uses new methods to do translations in whole sentences, rather than word-by-word or phrase-by-phrase.
Researchers have always known that the “whole sentence approach” is more effective. But the problem has been that, up until now, it has taken computers too long to process the information.
The rise of artificial intelligence and machine learning has made computers more nimble and enabled companies like Google to take advantage of the more effective technique.
India sees huge gains
This advance is especially important in the so-called developing world, where travel companies are seeing their fastest growth today.
India, for instance, has about 400 million Indian language users who are online, according to a Google survey. But given that about one out of five Indians doesn’t read English fluently, much of the internet remains inaccessible to many users, including travelers.
Google is applying its new translation technology to allowing people in India and elsewhere who don’t use one of the eight most popular languages in the world to read content online in their native tongue.
A month ago, Google was able to add 11 widely used Indian languages, including Hindi, Bengali, Marathi, Tamil, Telugu, Gujarati, Punjabi, Malayalam, and Kannada, to the 11 it already covered.
It says support for more dialects and languages in India and elsewhere are due this year in its search, its Translate apps, and its translate text webpage.
Google is racing to compete with its peers, such as Facebook and Microsoft, that are ramping up their skills at translating text, too.
Earlier this month, Facebook revealed it was also hard at work at deploying neural machine translation for its platform.
Microsoft is about to release a translation app based on the neural machine translation technique for Windows and other operating systems that can work with spoken language, via its Cortana voice-recognition service, to complement its services for text translation via Bing and Skype.
There’s money in this for the tech companies, of course.
For instance, Google is making its translation service available to companies at a charge of $20 per million characters translated. As of today, English to and from Chinese, French, German, Japanese, Korean, Portuguese, Spanish, Turkish, Russian, Hindi, Vietnamese, Polish, Arabic, Hebrew, and Thai are available using the new technique, with more on the way.
Smaller companies may try to give the big players a run for their money.
Naver, a Korean tech company, has used the same neural machine translation method to develop an app called Papago that is winning popularity over Google and other Western brands in some parts of Asia.
Skift Daily Newsletter
Get the travel industry’s daily must-read email 6 days a week
Photo credit: Using a new method, computers can now produce much more accurate translations. Anthonyboyd / Freepik