Event Page

Chaitanya Chokkareddy explores Small Language Models (SLMs) as an alternative to Large Language Models (LLMs) for supporting Indian languages. While large models offer powerful language capabilities, they often prioritize languages with extensive digital resources, leaving many Indian languages underrepresented. SLMs may provide a more practical and community-centered approach by requiring less data, computing power, and infrastructure.

The talk examines how smaller, specialized models can be trained on carefully curated linguistic resources, including community-approved texts, oral histories, dictionaries, and educational materials. Such models could support language documentation, translation, teaching, and conversational learning while allowing communities greater control over their linguistic and cultural data.

The discussion will also address challenges, including limited training resources, model accuracy, technical expertise, and the need for ethical data governance.