Bengaluru's Sarvam AI Launches Mayura to Improve English-to-Indic Language Translation

Bengaluru-based startup Sarvam AI has launched a new translation model called Mayura to address long-standing challenges in English-to-Indic language translation, particularly in everyday, colloquial communication.

The model focuses on real-world language patterns such as code-mixing and colloquial expressions, aiming to deliver translations that are more relatable and accurate for Indian multilingual speakers. It supports translations between 10 major Indian languages, including Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali.

Conventional translation models have struggled to capture the nuances of how Indians actually communicate, as they often mix regional languages with English. Sarvam Translate, however, was developed with a unique approach: it utilizes diverse real-world data to train the model, ensuring that regional dialects, slang, and even code-mixed phrases are preserved in translations. 

This development is set to enhance the accessibility of digital content, social media, and e-commerce services for millions across India.

The model also tackles gender-specific language in Indic languages, a critical area where traditional models fall short. By including a "gender toggle" for first-person translations, Sarvam Translate provides appropriate gender representation in conversations, both for AI-powered voice chatbots and human agents.

The application of this model extends beyond casual conversations. It’s also designed for domain-specific translations, like legal, scientific, and technical documents, ensuring that complex jargon is accurately translated into simple, accessible Indic language. Sarvam AI’s dual-stream architecture preserves formatting elements in technical content, making it especially useful for educational and government communications.

Mayura is now available as an API, allowing developers and businesses to integrate it into their applications. As part of Sarvam AI’s suite of products, it joins other tools such as Sarvam Agents, Sarvam 2B (an open-source language model), and Shuka 1.0 (an audio language model).