How regional language technology became flavour of the season
There are a clutch of homegrown companies which are trying to understand what Indian language speakers want to talk about and how to encourage them to do it.
The common thread connecting these companies is a focus on solutions that range from bots for local language banking telephony to social platforms that intuitively connect farmers to agricultural experts.
Gnani.ai, for instance, has developed products based on voice recognition technology in languages including Hindi, Kannada, Telugu, Tamil and Gujarati, and also for those who speak in English, but with an Indian accent.
The company, founded by Ananth Nagaraj and Ganesh Gopalan who were colleagues at Texas Instruments, is among three others that have received investments totaling Rs 60 crore from Samsung Venture Investment Corporation (SVIC), the venture capital arm of South Korean conglomerate Samsung Group.
Although Gnani.ai is currently focused on building AI-powered local language support for products that serve banks, financial institutions and farmers, the investment in its automatic speech recognition engine could potentially power Samsung’s multi-platform voice assistant Bixby in regional language services in future.
“If you’re from a small village in Haryana and want to know your account balance, you can ask that in Hindi and get an automated answer, either through the telephone or on an app,” said cofounder Ganesh Gopalan.
The challenge of getting quality local language support for basic transactions prompted the team at Gnani.ai to build speech-powered solutions. For banks with popular mobile banking apps, the company has built a voice assisted layer that sits on top of the existing app platform. Typically, these large nationalised banks and institutions serve multiple regions and cater to various languages.
The company is able to guarantee the accuracy of its speech recognition due to the technology team’s collaboration with linguists to constantly improve accuracy and context, Gopalan said. Since the technology allows people to ask questions in their own language and receive automated answers, its functionality has been extended to a state government project that was looking to help farmers.
“If a farmer wants to find the price of a particular commodity at a local mandi (market), we bridge that information gap in his own language with a number to call and ask. We have done some POCs (proofs of concept) in some languages to ask for the price of a particular crop at a particular mandi,” Gopalan said.
Applications for local language speech recognition also extend to large scale surveys by the central government. For instance, Voxta Communications, a Hyderabad and London-based speech technology company, is helping the National Skill Development Corporation (NSDC) understand citizen feedback on a range of issues.
“The content is immaterial — whether you are saying here’s how you get information about pre- or post-natal health or how to inoculate your child against chicken pox, you can hang different material on our mobile bot framework,” said cofounder Kavita Reddi.
The company, earlier dubbed ‘political Siri’ for its use in Prime Minister Narendra Modi’s 2014 election campaign, was recently recognised for work with FMCG major Reckitt Benckiser in a voicedriven Swachch Bharat campaign that drove awareness on personal hygiene and cleanliness.
Reckitt’s consumer hygiene brand Dettol was able to collect substantial data through speech recognition over phones on what inhabitants of remote villages most required for basic sanitation in their localities. Dettol was able to get users to interact with the brand and give their opinions, simply by ‘talking’ on the phone’, according to the company.
“The brand used Amitabh Bachchan’s voice to read a poem on health and hygiene. We did an outbound campaign for them for five million people. They gave people the choice on how to be healthy and hygienic or listen to more of the poem or to contribute their opinion...it turned out that most of the people wanted toilets in their villages,” said Reddi.
Voxta is now working with the NSDC to engineer regional language voice recognition and interaction systems, building on the success of earlier projects, including one around maternal health in women. The company’s speech system is expected to help the government conduct surveys around health and hygiene, with citizen feedback in multiple languages.
Reverie Language Technologies, acquired by Reliance Industries earlier this year for ?190 crore, has also built multiple solutions ranging from fonts for Indic scripts, and keyboards to content conversion.
Reverie is trying to help farmers make better sense of commodity pricing data in their own language and give easy-to-understand reports for their reference, as part of an ongoing government project on text transliteration for the National Agriculture Market (eNAM) portal. The project started in 2016 and is for a period of five years.
More than 4.5 million farmers across the country use the eNAM platform to help make decisions on selling their produce. Reverie has localized the eNAM portal’s content in five languages – Hindi, Marathi, Gujarati, Telugu and Bengali — with real-time transliteration and translation for static and dynamic content.
“We also enabled the generation of real-time per user reports in local languages in multiple formats, on a daily basis,” said Mithun Das, Chief Revenue Officer, Reverie Language Technologies. The company counts HDFC Bank, Practo, BHIM and Intex among its other clients.
Speech and text input solutions aside, there are also a clutch of homegrown companies which are trying to understand what Indian language speakers want to talk about and how to encourage them to do it.
Audio platform Vokal, for example, has found success in engaging people who ask questions in Indian languages with those who provide high-quality answers.
A quick look at the platform can throw up uniquely Indian questions like — “How do I prepare for an IAS exam if I am not good at English?” or “Is rice made in a pressure cooker bad for health?” — and provide equally eager answers to such queries.
Cofounder Aprameya Radhakrishna says the depth of discussion among local language users on basic topics such as education, wealth and politics is vastly different from the Englishspeaking audience.
“For example, if you go to the English-speaking world, they may ask ‘how do I invest better, how do I save up for my retirement’ whereas for the next billion it is about ‘how do I make money in the first place’, leave alone investment or buying insurance,” said Radhakrishna.
These key differences made the company believe that it was important to build a different product for non-English users that allowed for a strong sense of community and shared knowledge-building.
Sreeraman Thiagarajan, cofounder of Agrahyah Technologies which builds audio and text content in regional languages, said enabling meaningful and useful conversations — not just light-hearted chatter — in Indian languages is a worthwhile exercise for the next billion.
Thiagarajan holds equity in Mooshak, a ‘made-for-India’ platform for users to consume, create, and share content in Indian languages. On the Mooshak app, a conversation between grape farmers of Nagpur drove home the essence of building a platform for useful discussions in local languages.
Grape farmers in Nagpur were talking to each other about a new mutated variety of fungus that they could not combat. “The government does have agricultural scientists on call to solve such problems, but they don’t have the wherewithal to do it real-time or on a big scale — this was a shared purpose for people to discuss and figure out solutions,” he said.
The four-year-old app is undergoing a revamp and will re-launch in August. Following the entry of regional social content platform ShareChat, the app slid on the popularity charts. The revamp will, however, focus completely on facilitating people to find solutions to their problems.
“We cannot outshout a 200 million user base of TikTok (a popular short-video app), but we are building a meaningful, smaller platform,” Thiagarajan said.
To be sure, big technology companies have built technology enablers to easily access content in Indian languages. Google Go, for instance, lets users listen to web pages in 28 languages, including Hindi, Bengali, Malayalam, Marathi and Tamil. This can happen even on slow 2G connections through natural language processing and speech synthesis AI.
Microsoft recently released phonetic keyboards for 10 Indian languages during an update for its Windows 10 operating system in May. The virtual keyboard learns from a user’s behavior pattern and preferences, and offers individualized word suggestions in Indian languages - improving accuracy of text input, it said.
Amazon launched a new skill for Alexa in India last year, aimed at helping the platform learn Hindi and other languages like Tamil, Marathi, Kannada, Bengali, Gujarati and Telugu. The Indian companies, however, want to immediately solve the dayto-day problems of non-English users rather than wait for large technology companies to fix the gap.
“The whole of India is not going to learn English overnight; it is going to take decades. This is a window of opportunity where you can build a brand by being a great product in Indian languages,” Vokal’s Radhakrishna said.