If you think translations of your social media posts by Google and Facebook are rudimentary, here comes the 2.0 of machine translations, with an accuracy level of 80-90 per cent.

Kshanakalam Aanimutyam , a Telugu translation of S Sadakshari’s Kannada work ( Kshana Hottu Aanimuttu ), becomes the first book in India to be translated by the machine. And, Saara Machine Translation developed by the Hyderabad Central University takes just a second.

Instant translation

Using this solution, all of modern Kannada works can be translated into Telugu. No matter how big the work is, it will be translated into Telugu just in a second. It, however, needs vetting by a person who knows both the languages to finetune the grammar and syntax here and there.

The 150-page Kannada book was translated into Telugu in a pilot anchored by the Hyderabad Central University.

The process involves two steps. “All you need to do is post the text in the required area and press enter. Alternatively you can upload the files as well. You will get a transliterated version of the text that makes it easier for the computer to understand and translate,” said Kavi Narayana Murthy, who led the team at the School of Computer and Information System.

Showcasing the process, Narayana Murthy said the method can now be replicated to other languages.

Appa Rao Podile, Vice-Chancellor of HCU, claims that this is the first work in the India to be machine translated.

The difference between Saara translator and Google translator is, the latter is fully automatic whereas Saara is not fully automatic.

“Google requires billions of parallel sentences (translated sentence pairs) for training to get high quality automatic translations. Saara Translator does not require billions of sentences for getting the same result,” Ram Anirudh, a research scholar with the Saara team, said.

Human intervention

“We get high quality in Saara translator, by having a human in the loop. Over some time, machine performance improves and human effort required will be minimal,” he said.

The team took three years to develop the solution. It is now working on a solution that can translate the texts from Telugu into Kannada.

There is no need for a linguist to oversee the process. Any person who knows both the languages can vet the translated version to correct mistakes. “It might take 20 minutes a page for correction or rewriting,” he added.

comment COMMENT NOW