More than 400 million people in India use the internet, and more are coming online every day. But the vast majority of India’s online content is in English, which only 20 percent of the country’s population speaks—meaning most Indians have a hard time finding content and services in their language.
Building for everyone means first and foremost making things work in the languages people speak. That’s why we’ve now brought our new neural machine translation technology to translations between English and nine widely used Indian languages—Hindi, Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Malayalam and Kannada.
Neural machine translation translates full sentences at a time, instead of pieces of a sentence, using this broader context to help it figure out the most relevant translation. The result is higher-quality, more human sounding translations.
Just like it’s easier to learn a language when you already know a related language, our neural technology speaks each language better when it learns several at a time. For example, we have a whole lot more sample data for Hindi than its relatives Marathi and Bengali, but when we train them all together, the translations for all improve more than if we’d trained each individually.
These improvements to Google Translate in India join several other updates we announced at an event in New Delhi today, including neutral machine translation in Chrome and bringing the Rajpal & Sons Hindi dictionary online so it’s easier for Hindi speakers to find word meanings right in search results. All these improvements help make the web more useful for hundreds of millions of Indians, and bring them closer to benefiting from the full value of the internet.
Half the world’s webpages are in English, but less than 15 percent of the global population speaks it as a primary or secondary language. It’s no surprise that Chrome’s built-in Translate functionality is one of the most beloved Chrome features. Every day Chrome users translate more than 150 million webpages with just one click or tap.
Last year, Google Translate introduced neural machine translation, which uses deep neural networks to translate entire sentences, rather than just phrases, to figure out the most relevant translation. Since then we’ve been gradually making these improvements available for Chrome’s built-in translation for select language pairs. The result is higher-quality, full-page translations that are more accurate and easier to read.
Today, neural machine translation improvement is coming to Translate in Chrome for nine more language pairs. Neural machine translation will be used for most pages to and from English for Indonesian and eight Indian languages: Bengali, Gujarati, Kannada, Malayalam, Marathi, Punjabi, Tamil and Telugu. This means higher quality translations on pages containing everything from song lyrics to news articles to cricket discussions.
The addition of these nine languages brings the total number of languages enabled with neural machine translations in Chrome to more than 20. You can already translate to and from English for Chinese, French, German, Hebrew, Hindi, Japanese, Korean, Portuguese, Thai, Turkish, Vietnamese, and one-way from Spanish to English.
We’ll bring neural machine translation to even more languages in the future. Until then, learn more about enabling Translate in Chrome in our help center.
Just over a year ago, we saw a major milestone in the field of artificial intelligence: DeepMinds AlphaGo took on and defeated one of the worlds top Go players, the legendary Lee Sedol. Even then, we had no idea how this moment would affect the 3,000 year old game of Go and the growing global community of devotees to this beautiful board game.Read More…
One of the great promises of AI is its potential to help us unearth new knowledge in complex domains. Weve already seen exciting glimpses of this, when our algorithms found ways to dramatically improve energy use in data centres – as well as of course with our program AlphaGo.Read More…
Its now nearly a year since DeepMind made the decision to switch the entire research organisation to using TensorFlow (TF). Its proven to be a good choice – many of our models learn significantly faster, and the built-in features for distributed training have hugely simplified our code. Along the way, we found that the flexibility and adaptiveness of TF lends itself to building higher level frameworks for specific purposes, and weve written one for quickly building neural network modules with TF. We are actively developing this codebase, but what we have so far fits our research needs well, and were excited to announce that today we are open sourcing it. We call this framework Sonnet.Read More…
From binge-watching your favorite TV shows with Chromecast, to searching online for a cookie dough recipe for a night in, having great Wi-Fi at home helps with special everyday moments. But as we all know, sometimes these moments can turn into hours spent watching videos or browsing photos.
That’s why we built Scheduled Pause, a new feature in Google Wifi that lets you automatically pause the internet for everyday events like “Bedtime” to help wind down at the end of the day, or have a daily “Homework” schedule so your kids can focus before dinner.
The idea behind Scheduled Pause started a year ago. While exploring how to best create tools for families, I noticed that I was having trouble falling asleep. I’d check emails and surf the web late into the night. Experimenting with options, I started using a timer on my computer to turn the internet off at 11 p.m. The first night was a shock, but after a few nights I was ready to shut down earlier. And I was more refreshed and rejuvenated in the morning.
As I started talking to more people, in and outside of Google, I found that screen time was a common challenge for parents—from getting kids to put down their favorite game to struggling to have dinner without eyes glued to devices.
We hope Scheduled Pause helps you and your family create time for everyone to be more present and enjoy everyday moments.
We’re honored to have partnered with Dr. Louise Banks, esteemed linguistics professor, to develop instant camera translation for our 32nd language, Heptapod B. Following our experience with logograms in Chinese and Japanese, as well as the many characters containing circles in Korean, we were ready to blend our expertise in low-memory-footprint convolutional modeling and Dr. Banks’ linguistic background to the deciphering of circular logograms in the Word Lens feature in the Google Translate app.
The challenge with understanding Heptapod B is its nonlinear orthography. Fortunately, Google’s neural machine translation system employs an encoder/decoder system that internally represents sentences as high-dimensional vectors. These vectors map well to the non-linear orthography of the Heptapod language and they are really the enabling technical factor in translating Heptapod B.
We interpret Heptapod B into English, Chinese, Danish, Japanese, Urdu, Russian, French, Spanish and Arabic. As with our other Word Lens languages, it works offline, which is really handy if you happen to need to read a circular logogram in an isolated location. Dr. Banks assures us that the app will continue to work for at least 3,000 years.
Communicating across language (and glass) barriers can be a rather alienating experience. While learning a new writing system can be quite rewarding and even a mind-altering experience, not everyone has time for that. So whether the world’s fate hangs in the balance, or if you’re simply trying to discern whether your coffee stain ring means something, we wish you success as you integrate this tool into the story of your life.
(Okay, if you haven’t guessed already… we’re just having some fun here. But we really are eager to bring Word Lens and Neural translation to more languages,
so stay tuned.)