How to operate AI in its 22 languages

Vineet Sawant has spent the last two years navigating the streets of Mumbai on a scooter as a delivery driver.

“Being on the road is always very stressful and especially in cities like Mumbai,” he says.

But when he started the linguistic barriers was an additional problem.

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

This caused problems with his new job.

He said: “At first it was difficult. Everything was in English, and I can understand a game, but I am more comfortable in Marathi. I asked other deliveries to help me understand what to do.”

His employer, Zepto, promises “the fastest online grocery delivery of India”. So having drivers in the grip with delivery instructions was not ideal.

To smooth this process a year ago, Zepto joined forces with Reverie Language Technologies to introduce an AI translation service for its pilots.

Since then, its delivery pilots have been able to choose between six languages on the Zepto application.

“I no longer have to guess,” said Mr. Sawant.

“Earlier, I took more time to read and sometimes even make mistakes. Now, if the customer writes” Ring Bell “, I receive this instruction in marathi. So, I don’t have to ask or check again. Everything is clear.”

Mr. Sawant’s difficulties are common.

“India has 22 official languages and hundreds of dialects,” said Professor Pushpak Bhattacharyya, of Iit Mumbai, one of the main experts in India in the use of AI in Indian languages.

“Without technology, which understands and speaks these languages, millions of people are excluded from the digital revolution-in particular in education, governance, health care and banking services,” he said.

The deployment of new generative AI systems, such as Chatgpt, has made the task more urgent.

Vast amounts of data, such as web pages, books or video transcriptions are used to form an AI.

In largely spoken languages like Hindi and English, it is relatively easy to obtain, but for others, it is more difficult.

“The main challenge for creating Indian language models is the availability of data. I am talking about refined data. Gross quality data is available. But this data is not very high quality, it needs filtering,” said Professor Bhattacharyya.

“The problem in India is for many Indian languages, in particular tribal and regional dialects, these data simply do not exist or is not digitized.”

Reverie Language Technologies is currently deploying its IA -managed translation technology for a range of Indian companies.

The Vivekananda Pani co-founder says that if translation technology will facilitate communication, there is “potential for less common dialects to be dismissed”.

“The challenge will be to ensure that the incredible advantages of Linguistic Progress focused on AI do not accidentally reduce the rich variety of human language.”

To help solve the problem, Professor Bhattacharyya contributed to Bhashini, a government project to develop these high quality data sets necessary to form an AI.

In addition to data sets, Bhashini has built AI language models and 22 languages translation services.

Started in 2022, it is a huge company, but has already made a lot of progress.

Bhashini is currently hosting 350 language models based on AI which have treated more than a billion tasks.

More than 50 government services work with Bhashini, as well as 25 governments of states.

For example, Bhashini Tech is used in multilingual chatbots for public services and to translate government programs in local languages.

“Bhashini ensures a linguistic and cultural representation of India by building IA models specific to India rather than relying on global platforms”, explains Amitabh Nag, CEO of Digital India, Bhashini Division.

He hopes that over the next two or three years, rural users will have access to the voice to government services, financial tools and information systems in their mother tongue.

These sets of data focused on India, hopefully, will one day give people to develop models based on AI The tools to facilitate adaptations for the entire population.

Currently, the design of any AI program to manage complex processes such as health care can be extremely difficult.

Kshitij Jadhav, an associate professor at the Koita Center for Digital Health at Iit Mumbai, works on an AI program that would help people quit smoking.

He explains that people at different stages of the process need different advice and that they generally need a well -trained human to do this assessment.

But there are a limited number of practitioners who can help, in particular those who can operate in several languages, therefore Professor Jadhav hopes that his IA model can fill the gap.

The AI “will first identify the type of conversation that the person needs and will therefore supervise the questions, will show empathy, emotions,” said Professor Jadhav.

And all of this, hopefully, will ultimately be done in 22 languages. The first experiences are underway in English and Hindi.

“It will be very personalized, it will not be something right next to the shelf,” he said.

Back in the streets of the city, Vineet Sawant increased the number of packages that it offers from ten to around 30 per day, helped partly by the translation function in the delivery application.

He thinks it will help more people like him.

“It gives us the impression of belonging. Everyone does not understand English. When the application speaks our language, we feel more confident and we work better.”