While online content in India is dominated by English, only about 20 percent of the country's population speak that language. Google, therefore, sees a disconnect between the kind of online content Indians seek in local languages and what they usually end up getting.
A potential solution, the search giant thinks, would be a platform allowing much better translation to and from English to make "things work in the languages people speak." To make that happen, Google announced on Tuesday that its neural machine translation technology will now help people translate between English and nine widely used Indian languages, including Hindi, Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Malayalam and Kannada.
These nine Indian languages are part of more than 20 other languages enabled with neural machine translations in the company's Chrome browser. Google said that it would include even more languages to the platform in the future. But, what is neural machine translation, and how does it help people get better translations?
Introduced in November 2016, neural machine translation was the next step Google took to improve its Google Translate tool after using statistical models to translate text. What sets it apart from statistical machine translation is its ability to translate complete sentences at a time, instead of pieces of a sentence.
Focusing on a broader context helps the technology provide the best possible translation, which is "more human sounding." According to Google:
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.
Google previously used a phrase-based machine translation that used to break a sentence into words and phrases to be translated largely independently. The neural machine translation, on the other hand, works on the entire sentence as a single unit for translation.
For successfully translating an entire sentence, the new neural machine translation technology first rearranges it and then adjusts it to sound more like a human, who has proper grammatical skills. The technology is developed in a way that it learns over time to offer translations that are smoother and easier to read.
As part of a study, researchers at Google tested the company's neural machine translation system by using a human side-by-side evaluation on a set of simple sentences. They found that the technology could reduce translation errors by an average of 60 percent compared to the previously used phrase-based production system.
In addition to bringing neural machine translation support to additional Indian languages, Google also made the popular Rajpal & Sons Hindi dictionary available online to make it more convenient for Hindi speakers to find word meanings on the web.
