Google Translate is fast developing into one of those universal translators that we only see in science fiction, most notably, Star Trek. The latest update to the platform brings improved support for 9 different Indian languages via its new, neural-network based translation tools.
Traditionally, Google Translate, and other language translation services, worked on what is known as a phrase-based translation algorithm. This is still better than a word-based translation mechanism as a phrase-based one translates commonly used sentences very well.
However, such a mechanism is still not a very intelligent one and doesn’t appreciate context, especially in complex sentences.
Google’s solution to this is the Google Neural Machine Translation system (GNMT). If you’re interested in the technical details, head to Google’s research paper here.
Essentially, GNMT learns to map the input sentence to an output one as a whole rather than break it down into bits.
This new technique, says Google, brings down translation errors by “more than 55 percent to 85 percent.” The end result is something that looks very similar to a human translated sentence, claims Google.
Using this technique, Google has been able to improve translations for 9 different Indian languages including Hindi, Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Malayalam and Kannada.
GNMT works best with a large data set; the more the data, the better the translation. As Google explains, there is a large data set available for Hindi, but not for other languages. This was a problem, but Google apparently discovered that when the translations were trained together, all translations improved overall.
Neural machine translations are only one of the many updates that Google announced, however. The same abilities are coming to Google Chrome and search, enabling you to view better translations of web pages directly in the browser.
The Rajpal & Sons Hindi dictionary is also available online, so Hindi word meanings will now be easier to find.
Google points out that over 150 million pages are translated to Indian languages every day.
Published Date: Apr 26, 2017 12:20 pm | Updated Date: Apr 26, 2017 12:20 pm