How to force AI to work in 22 languages

The last two years Vineet Sawant spent, moving through the streets of Mumbai on the scooter as a shipping driver.

“It is always very tense on the road, and especially in cities such as Mumbai,” he says.

But when he started language barriers, he was an additional problem.

His first language is Marat, and Mr. Sont speaks “very little” English. “I can understand, but it’s very difficult to read,” he explains.

This caused problems with his new work.

He said, “It was hard at first. Everything was in English and I could understand some of them, but I am more comfortable in the Marathi. I asked for other shipping guys to help me find out what to do.”

His employer, Zept, promises “the fastest delivery of grocery products in India”. Thus, drivers fighting the delivery instructions were not perfect.

To smooth this process a year ago, Zepto collaborated with Reperie Mangual Technologies to submit AI translation service for its drivers.

Since then, the shipping drivers have been able to choose between six languages on the ZEPTO app.

“I no longer have to guess,” says Mr. Savant.

“I used to take more time to read and sometimes make a mistake. Now that the client writes” Call “, I get this instruction in the Marathi. So, I don’t need to ask or check again. Everything is clear.”

Mr. Savanta’s difficulties are common.

“India has 22 official languages and hundreds of dialects,” says Professor Pushpak Bhattacheria from IIT Mumbai, one of the leading Indian experts on Indian.

“Without technologies that understands and speaks these languages, millions are excluded from the digital revolution – especially in the field of education, management, health and banking,” he emphasizes.

The release of new AI generative systems, like Chatgpt, has made the task more relevant.

A lot of data such as web page, books or video stents are used for training.

In wide spoken languages such as Hindi and English, it’s relatively easy to get, but it is more difficult for others.

“The main problem of creating Indian is the availability of data. I am talking about exquisite data. The data is coarse. But these data are not very high quality, it needs filtering,” says Professor Bhattacheria.

“The problem in India lies in many Indian languages, especially tribal and regional dialects. These data simply do not exist or are not digitized.”

Now Reverie Mangea Technologies unfolds its AI translation technology for various Indian companies.

Vivekananda Pani co -founder says while translation technology will facilitate communication, there is “the potential for less common dialects that will be repelled.”

“The task is to make sure that the strange benefits of promoting AI language are not accidentally reduced by the rich variety of human language.”

To help solve the problem, Professor Bhattacharia Made in the bhadznoA government project to develop these quality data sets required for training II.

Like data sets, Bhashini created language models AI and translation services in 22 languages.

It began in 2022, this is a huge endeavor, but it has already made a lot of progress.

Currently, Bhashini conducts 350 AI -based language models that have processed more than a billion tasks.

More than 50 government agencies work from Bharodni, as well as 25 state governments.

For example, Bhashini Tech is used in multilingual chats for public services and to translate public schemes into local languages.

“The Bhasniki guarantees the linguistic and cultural representation of India, creating Indian models of AI, not relying on global platforms,” says AMITAB NAG, CEO of Digital India, Bharosnia Division.

He hopes that over the next two or three years, rural users will gain access to public services, financial instruments and information systems in their native languages.

Hopefully, these sets of India data will give people who develop AI -based models, tools to greatly alleviate their adaptation to the entire population.

Currently, the development of any AI program to combat complex processes such as healthcare can be extremely complicated.

Kshitij Jadhav, Associate Professor at the Koita digital health center in IIT Mumbai, is working on AI program that will help people quit.

He explains that people at different stages of the process need different tips, and usually a well -prepared person is needed to make this assessment.

But there are a limited number of practitioners that can help, especially those that can work in several languages, so Professor Jadkhov hopes that his AI model can overcome the gap.

Ai “first determine the type of conversation that a person needs, and accordingly, they will make questions, show empathy, emotions,” says Professor Jadkhov.

And all this, hopefully, will eventually be done in 22 languages. Initial experiments are conducted in English and Hindi.

“It will be very tuned, it will not be something near the shelf,” he says.

Returning to the city street, Vineet Sawant has increased the number of parcels that it delivers from ten to 30 a day, partially helped the translation function in the delivery app.

He thinks it will help more people like him.

“It makes us feel what we belong. Not everyone understands English. When the application speaks our language, we feel more confident and we work better.”