Hindi → Indian Languages Governance Translation Dataset

Overview

Parallel corpus for Hindi → multiple Indian languages focused on the governance/citizen-services domain.
Content includes government notifications, schemes and benefits, debates from Lok Sabha and Rajya Sabha, PM speeches etc. Each subfolder corresponds to a language pair (e.g., HIN-ASM, HIN-BAN, HIN-DOI/ODI, HIN-GOM, HIN-GUJ, HIN-KAS, HIN-MAI, HIN-MAR, HIN-MNI, HIN-NPI, HIN-ODI, HIN-PAN, HIN-SAT, HIN-SND, HIN-TEL; plus Tamil-centric TAM-KAN, TAM-MAL, TAM-TEL if present).

Languages

Source: Hindi (hi)
Targets: Assamese (as), Bangla (bn), Goan Konkani (gom), Gujarati (gu), Kashmiri (ks), Maithili (mai), Marathi (mr), Manipuri/Meitei (mni), Nepali (npi), Odia (or), Punjabi (pa), Santali (sat), Sindhi (sd), Tamil (ta), Telugu (te), Kannada (kn), Malayalam (ml), Urdu (ur).

Sources

Replace with exact attributions:

• Central/state government portals and circulars (publicly available)
• Citizen charter pages, scheme brochures, and helpdesk FAQs
• e-Governance service portals and SMS/IVRS templates
• Verified crowdsourced translations with reviewer QA

Format

• Folder arrangement: Each language pair is subdivided into sub-domains and each sub-domain is further divided into source-translated, source_reviewed and translated-reviewed language pairs.
• Type folders: source-reviewed translations are the translations of the highest quality which are reviewed after translation; whereas source-translated language pairs are larger in number but are unreviewed after manual translation. The difference between the qualities can be estimated using the translated-reviewed language pairs using TER.
• File type: TSV per language pair
• Fields: id, src_hi, tgt_<lang>, domain="governance"
• UTF-8 encoding, normalized punctuation

Intended Use

• Training & evaluation of MT systems for governmental content in India.