In a bustling city like Mumbai, navigating the streets can be a daunting task—especially for delivery drivers facing the dual challenge of traffic and language barriers. Vineet Sawant, a delivery driver for Zepto, has experienced these challenges firsthand. Over the past two years, he has taken to the roads on a scooter, grappling not only with stressful delivery routes but also with language difficulties. As a native Marathi speaker who struggles with English, Mr. Sawant initially found it hard to comprehend delivery instructions, often relying on fellow drivers for assistance.
Recognizing the critical need for effective communication among their drivers, Zepto partnered a year ago with Reverie Language Technologies. This collaboration led to the implementation of an AI-powered translation feature in the delivery app, allowing drivers like Mr. Sawant to choose from multiple languages that include their mother tongue. Now, he can instantly receive instructions in Marathi, which significantly reduces misunderstanding and increases his efficiency, allowing him to boost his daily delivery count from ten to around 30 parcels.
Language diversity is a major issue in India, which boasts 22 official languages and numerous dialects. "Without technology that understands and communicates in these languages, many are excluded from the benefits of the digital age," states Professor Pushpak Bhattacharyya from IIT Mumbai. He highlights that the availability and quality of training data for AI in less commonly spoken languages remain significant hurdles. While prominent languages like Hindi and English have ample resources, many regional dialects suffer from a scarcity of high-quality, digitized data.
Reverie Language Technologies aims to address these challenges, but concerns arise regarding the potential sidelining of less common dialects. Co-founder Vivekananda Pani emphasizes the importance of preserving linguistic diversity even as AI translation advances. Alongside industry efforts, initiatives like Bhashini, launched by the government in 2022, are making substantial strides in this space. Bhashini is committed to developing superior datasets and AI models for various Indian languages, facilitating tasks for over 50 government departments.
Just over a year into its operation, Bhashini hosts 350 AI language models and has processed over a billion tasks. From multilingual chatbots for public services to translating government initiatives, the program aims to provide localized access to information that respects India’s linguistic heritage. Amitabh Nag, CEO of Digital India’s Bhashini division, envisions a future where rural users can access critical services in their native languages within the next few years.
Research efforts in the healthcare sector are also underway. Professor Kshitij Jadhav, at IIT Mumbai, is developing AI solutions to help individuals quit smoking. The AI will tailor advice based on the user's specific needs, trying to address communication barriers as it progresses towards supporting 22 languages.
Back in Mumbai, Vineet Sawant observes a noticeable change in his confidence and ability to connect with customers. The AI translation feature not only facilitates better communication but also fosters a sense of belonging for drivers operating in a linguistically diverse society. "When the app speaks our language, we feel more confident, and we work better," he reflects.
