Articul8 develops AI models rooted in Indian language heritage
Summary and headline written by AI from the source article. How we work
In India, Articul8 is building artificial intelligence models, Laya for Sanskrit and Sol for Tamil, that prioritize linguistic accuracy and cultural context.
The company’s Laya model demonstrated its precision when it disagreed with Sanskrit scholars offering an incorrect interpretation of a verse, relying on the ancient grammatical rules of Pāṇini. Union Finance Minister Nirmala Sitharaman unveiled Laya at Madras Sanskrit College on September 15, and the model is currently undergoing extended testing with scholars before wider release.
Articul8’s approach differs from many large language models by designing its AI to refuse incorrect answers and maintain a consistent understanding based on established knowledge. Arun Subramaniyan, CEO of Articul8, explained that general-purpose models like ChatGPT would have accepted the flawed interpretation. Laya’s architecture includes a custom tokeniser for Indic languages, a query breakdown model, a context embedding model, a rule verification engine, and the Laya generation model itself.
The company is also developing Sol, a Tamil language model using the framework successfully applied to Sanskrit. Articul8 is collaborating with institutions like Madras Sanskrit College, Banaras Hindu University, and the Pune Institute of Linguistic Sciences, and plans to expand the models to Gujarati, Kannada, and Telugu. The models may unlock insights from ancient texts related to fields like astronomy and metallurgy, potentially leading to new discoveries and commercial applications.
