IIT Patna launches COIL-D to build Indian language corpora

Per the TIH IIT Patna call for technology development, the Centre of Indian Language Data (COIL-D) project has an MoU with TIH IIT Patna and is soliciting expressions of interest to build language datasets and related services. The TIH notice lists deliverables that include corpora (text and voice) development, parallel corpora, and technology innovation for text processing and speech recognition. The TIH document states the project aims to improve translations from Hindi to 17 Indian languages and Tamil to 3 Dravidian languages, and identifies the initiative as sponsored by MeitY. The Bhashini project pages identify Dr. Asif Ekbal (IIT Patna) as the COIL-D chief investigator. The TIH announcement invited startups and corporates to participate and cited a first-cycle expression-of-interest deadline of 15 March 2025.

COIL-D is presented in public materials as a centralized repository and resource-creation effort for Indian languages, with stated deliverables focused on both monolingual and parallel corpora. Industry-pattern observations: large, government-backed corpora projects typically emphasize annotation standards, licensing clarity, and benchmarked evaluation sets to maximize downstream utility for multilingual models and machine translation. For practitioners, harmonized corpora and aligned parallel datasets reduce duplicated effort and can materially accelerate training of translation and ASR systems for low-resource Indian languages.

Public reporting and government platforms place COIL-D within the broader Bhashini ecosystem, which the Government of India frames as a national language technology initiative. Industry context: national-language repositories backed by ministries often open access to vetted datasets and create procurement pathways for startups and research groups, increasing data availability for academic and commercial model development.

Observers should track formal data licensing terms, annotation schemas, and the scope of languages and dialects included. Also watch for published benchmarks or leaderboards that would enable apples-to-apples comparisons across models using COIL-D resources. If TIH or Bhashini publish data access APIs or tooling, that will affect how practitioners integrate the resources into training and evaluation pipelines.