Vineet Sawant has been delivering groceries across the streets of Mumbai while facing the often overwhelming challenge of language barriers. The delivery driver, who primarily speaks Marathi and only has a limited command of English, recalls the difficulties he faced in understanding instructions at work. "In the beginning, I struggled with the English text," he admitted. With his employer, Zepto, striving to ensure "India's Fastest Online Grocery Delivery", these language issues were not ideal.
China's unique linguistic landscape, however, left many without necessary support as Sawant often turned to fellow drivers for help. To address this, Zepto teamed up with Reverie Language Technologies over a year ago to implement an AI translation service, allowing drivers to switch the Zepto app interface to six different languages. "Now I understand exactly what the customers want," Sawant expressed, stating that now he can interpret delivery instructions clearly without needing further assistance.
India is home to 22 official languages and an array of dialects, a challenge highlighted by Professor Pushpak Bhattacharyya from IIT Mumbai. He underscores the necessity of advancing technology that caters to these languages, especially since millions risk exclusion from essential services in education, governance, and healthcare. As generative AI systems like ChatGPT evolve, the rush to develop Indian language models becomes critical.
A significant roadblock in this transition is the scarcity of high-quality training data for less commonly spoken languages. Professor Bhattacharyya notes that while raw data exists, there’s a need for refined information, a particular challenge impacting tribal and regional dialects. Nonetheless, Reverie Language Technologies is extending its AI-driven translation capabilities across several Indian businesses to tackle this issue.
To bridge the gap, Bhattacharyya has been working with the Bhashini initiative, a government project aimed at curating necessary datasets to train AI systems. Launched in 2022, Bhashini aims to embrace the linguistic diversity of India, with 350 AI-based language models developed to date processing over a billion tasks. It collaborates with numerous government departments, providing critical services like multilingual chatbots and translations for local schemes.
Amitabh Nag, CEO of Digital India’s Bhashini Division, emphasizes the importance of these initiatives in ensuring cultural representation. He anticipates that rural users will soon access government services in their native languages, making technology more equitable.
As Indian researchers develop AI-driven solutions for various sectors, such as a smoking cessation model, the challenge remains for these innovations to be accessible in multiple languages, expanding their reach.
For Vineet Sawant, the translation feature has already made a profound impact on his productivity, increasing his daily parcel deliveries. He expressed hope that this technology will empower many more individuals faced with similar language barriers, stating, "We feel like we belong when the app speaks our language."
