Microsoft leverages AI to improve real-time translation

On the eve of India’s 69th Republic Day, Microsoft announced that it will be bringing its artificial intelligence (AI) prowess to real-time translation. This will help improve translations for Hindi, Bengali and Tamil. With the inclusion of deep neural networks, real-time translation for these regional languages will be more accurate thereby sound more natural.

Microsoft is not restricting translation of these languages to any specific app or service. AI will now help in more accurate translations across all of Microsoft be it the Edge browser, Bing search, Office 365 products like Word, Excel and PowerPoint among others including their dedicated Translator app for Android and iOS.

The Redmond giant will also release APIs on Azure for developers and partners so that they can incorporate this additional advantage into their products and services.

During the announcement, Sundar Srinivasan, General Manager – AI & Research, Microsoft India said, "We have supported Indian languages in computing for over two decades, and more recently have made significant strides on voice based access and machine translation across languages. Today’s launch is a testament of our quest to bring cutting edge machine learning tech to democratize access to information for everyone in India."

Microsoft details its advancements in the field of AI and neural networks further by claiming that it has been the oldest proponent to translate global as well as Indian languages through Statistical Machine Translation. 

In order to make the translation sound more natural, Microsoft’s deep neural network incorporates some theories about how multilingual humans perceive patterns and its effect. Another deep neural networks system TrueText, filters repetition, pauses, and indifferent words from a conversation which in turn enhances the appropriateness of a contextual translation.

"In India, about 12% of the people can speak, read, and write English, even though it’s not their first language. There are 600 million literate people, who aren’t necessarily proficient in English and prefer to consume information in their mother tongue," explains Krishna Doss Mohan, Senior Program Manager, Microsoft India.

Despite the complexities of Indian languages where our 29 states have 22 official languages, deep neural networks have shown at least 20% improvement in the output quality of the translation for Indic languages.

Siddharth Chauhan
Siddharth Chauhan is the Consumer Technology Reporter at Digit India. He used to work as an Assistant Editor at TechRadar India