Google has launched its keyboard Gbroad in 22 Indian languages, a move which will allow users to message in their native language. The languages now supported include languages like Assamese, Bengali, Bodo, Dogri, Gujarati, Hindi, Kannada, Kashmiri, Konkani, Maithili, Malayalam, Manipuri, Marathi, Nepali, Odia, Punjabi, Sanskrit, Santhali, Sindhi, Tamil, Telugu and Urdu.
The keyboard supports both language keyboards and transliteration into native languages. Apart from this, another major develop is that Google brought gesture/swiping based typing to Gboard for Indian languages: This was missing earlier. The keyboard also allows emoji and gif search using Indian language text. Apart from this, Gboard now allows the resizing of the keyboard for users. It also allows users to search the web via the keyboard, share content with friends within the app.
Expect apps like WhatsApp in particular to benefit from these improvements, more than Google’s own messed up portfolio of messaging apps.
A sign of things to come: Neural Network translation in 9 Indian languages
Google translate has been included in the Gboard, but clearly, the more significant development than Gboard is the launch of neural network translation in 9 Indian languages: Hindi, Bengali, Punjabi, Marathi, Gujarati, Tamil, Telugu, Malayalam and Kannada.
This expands neural network translation for Google beyond Hindi, allowing for far more efficient and accurate translation of Indian language text. Melvin Johnson, Engineer at Google Translate pointed out that even with 300 million Hindi speaking users online, Hindi isn’t among the top 10 languages online in terms of content. Google is addressing that with translation, including translation of pages.
“We perform 1 billion translations every single day”, he said, which amounts to 140 billion words, or 1 million books. Google has 500 million users globally using translate, and 95% of users are outside the US, with Brazil, India, Russia and Indonesia being the top countries. Johnson said that they’ve used neutal machine translation to bridge the gap between phrase based (less efficient) and human (more efficient) translation. “Neural machine translation allows the translation of the entire sentence, instead of on a piecemeal basis.” Google is switching all languages into a new system, and Johnson said that because of the efficacy of neural network translation, they’ll be able to launch in new languages.
Google builds language models, which allow translation from one language to the other, and with neural network processing, they’re able to translate without building one-to-one models. How does Google build these models? “Google has access to the entire web. We look for parallel documents on the web, which are trying to say the same thing in two different languages. For example, the BBC. We break it down and feed in examples of sentences in one language and in another, and it learns to do this mapping.”
“With neural systems, it takes 2-3 weeks to train per model, for hundreds for FPUs. For it to work really well, it needs hundreds of millions of examples for it to work really well. ”
Why this matters is that with time, this language translation capability will extend to voice and across devices, which is where the company is going with its Google Assistant.