Prime Minister Narendra Modi recently delivered a speech in Hindi that was translated in real time in Tamil through the sophisticated digital system called Bhashini. Developed by the Government of India, Bhashini has the mandate to build a National Public Digital Platform for languages that will use the power of AI and other emerging technologies to provide not only communication and translation related services but also diverse kinds of transaction related services to speakers of different Indian languages. Amitabh Nag, Chief Executive Officer, Digital India Bhashini Division, in conversation with Anoop Verma, Editor (Desk), ETGovernment, sheds light on the potential of Bhashini to break the barriers between different Indian languages by using emerging technologies.
Edited excerpts:
Why is an advanced language-translation system like Bhashini required in India?
India is a multilingual country. The country is home to more than 100 languages which are spoken by over hundred thousand people. Bhashini is basically a digital tool, based on Artificial Intelligence (AI), that enables real-time translation of Indian languages. By providing the facility of real-time translation, Bhashini makes it easy for Indians from different parts of the country to communicate with each other and conduct transactions with each other. Along with facilitating communication between people of different language groups, Bhashini fosters social inclusion by enabling access to the Internet and all kinds of digital services in local Indian languages. There was a latent need for a Bhashini like system because of India’s huge linguistic diversity.
The native languages in India have their own nuances related to grammar, pronunciation and other factors. How does Bhashini deal with these nuances?
To build Bhashini, we adopted various language formulas which are designed to deal with the nuances and constraints. The advancement that the world has made in the areas of GPU and AI can be seen as the fountainhead for an advanced system like Bhashini which ensures collaboration between people across the geographies and across the domains without getting a third party intermediary into the picture. Bhashini has been making steady progress in terms of the quality of its translations, and, over a period of time, we can expect it to deliver fully accurate results. You said that Bhashini is built on emerging technologies like AI. What is the technological architecture of Bhashini?
The technological architecture consists of multiple layers. We are using AI technology at the bedrock of this mission. The AI systems are essentially as good as the quality of data that is fed to them. The data can be put into two categories: what you are having and what you should be having. Currently, the internet is populated mostly with English content; the rest of the languages of the world are not adequately represented on the Internet. In the case of Indian languages, except for six or seven languages, the data does not exist for the rest of the languages as far as digital form is concerned.
One of the tasks for the Bhashini mission was to get the digital data created. For that we had 200+ translators in the field who were actually creating digital data by traveling to different parts of the country and speaking to the people and asking them to write something and speak in their own tongue. This is the process that was followed to create the corpus which was used to build the model. Bhashini is designed to be used by people of various linguistic backgrounds so it has to have the diversity that is required.
Another factor that we took care of while building our data is that language change is based on gender, age, profession and other aspects of life. There is a different flavor of the language that different segments of society use. By collecting our samples from different social verticals, we have enhanced the scope of our model. The creation of the digital linguistic data for Bhashini was a unique exercise–India is the only country to build this kind of project on a national level.
What are the processes that you followed for training and deploying the AI systems that lie at the core of Bhashini?
The AI models have been developed by around 70 research institutes. We began by dividing the problem into multiple projects, eight of them. Each of the eight projects were executed by a different consortium. The 70 research institutions were not dealing with the projects on an individual basis, they were dealing as consortium based on the functionality. For instance, there was the speech consortium and there were several other consortiums. As far as the deployment of the AI models is concerned, these models were supposed to be available as APIs. This platform has to be very strong and innovative.
There are two major stakeholders or users of the Bhashini system: the first is the service providers from the government and the private sector, and the second is the consumers. What kind of response are you getting from industry and the people?
The response is enthusiastic. Due to India’s linguistic diversity, there was a latent need for this kind of system. The Bhashini app has become very popular—it has got over 5,00,000 downloads so far. There are a number of use cases in which Bhashini can play a seminal role in enabling seamless transactions between people of different linguistic backgrounds. We are in touch with various organizations for implementing these use cases. Bhashini is being deployed for various tasks ranging from online chats, online document translation, voice translation and enabling voice based multilingual transactions.
UPI has become one of the most popular online payment systems in India. What kind of impact is Bhashini having on UPI and other such cashless payment systems?
We have already launched a voice based payment system that allows you to execute a person to person fund transfer using voice based commands. This service is available in a multilingual format. We will soon be coming out with a more extensive menu for conducting other kinds of digital transactions through Bhashini. By enabling voice based transactions, Bhashini is furthering the vision of social inclusion by enabling people who cannot read and write to conduct transactions without having to go through an intermediary. For the banks, and other institutions the use of voice based transactions means a faster turnaround time. Much faster and much safer processing of transactions is possible through voice based systems.
What kind of response are you seeing from masses in different parts of India? Are people using Bhashini’s services in large numbers?
The best indicator of the popularity of Bhashini is that our app, available in both iOS and Android, has been downloaded over 5,00,000. People are using the Bhashini app. Our web service app called Anuvad is also being used. Till about six months back, we used to have 20,000 to 30,000 inferences in a month, today we have about 4 crore inferences in a month. This is a huge increase in terms of usage. Bhashini is growing at a healthy pace because this system helps users conduct transactions in their mother tongue–there was always a need for this kind of system in India. Bhashini is collaborating with the ONDC platform to enable people to speak in their own local language while negotiating with people who speak other Indian languages. In future, I expect all e-commerce platforms to include systems which accept voice based instructions for executing transactions.
