SUVAS AI Translation: Supreme Court Tool Expands to 12 Languages as Courts Translate 50,000 Judgments
Judgments in regional languages are now available within 48 hours of pronouncement in pilot High Courts.
Key points
- The same-week translation pipeline — a draft from SUVAS, checked by a registry translator, published within 48 hours of pronouncement — now runs in twelve languages, up from the four it began with.
- High Court SUVAS cells have translated more than 50,000 judgments this year, on top of the Supreme Court's archive of over 50,000 translated judgments on e-SCR.
- Translations carry a disclaimer; the English text remains the authoritative version.
NEW DELHI — The Supreme Court's AI-assisted translation system now produces reviewed translations of new judgments in twelve Indian languages within 48 hours of pronouncement, the court's registry said this week, as High Court translation cells reported that their combined output this year had crossed 50,000 judgments. The same-week pipeline, which began with Hindi, Tamil, Gujarati and Odia, has added Bengali, Marathi, Telugu, Kannada, Malayalam, Punjabi, Assamese and Urdu, and the court has set a target of covering all 22 Scheduled languages.
"A judgment nobody can read is not a judgment anyone can follow," a member of the Supreme Court's committee on AI-assisted legal translation told The Counsel.
What SUVAS is
The Supreme Court Vidhik Anuvaad Software was launched in November 2019 as a machine-translation tool trained on legal text, producing drafts in nine languages that translators corrected before publication. Its early years were slow: a few hundred judgments were translated in the first year and fewer still during the pandemic. The project was reorganised in 2023 under a committee of the court, a backlog programme translated the court's reported judgments in bulk, and by mid-2024 the court told Parliament that more than 51,000 judgments had been translated into 18 languages and published on the e-SCR portal. High Courts set up their own SUVAS cells from 2023, translating their own judgments and the Supreme Court judgments arising from their states; the Allahabad High Court alone has published more than a lakh judgments in Hindi.
What has changed
- Speed. The archive programme translated old judgments months or years after pronouncement. The new pipeline translates each reportable judgment as it is pronounced: the English text goes to SUVAS on the day of pronouncement, a draft returns within hours, a registry translator checks it against the English, and the translation is published alongside the original within 48 hours.
- Languages. Twelve languages are now in the pipeline. The remaining Scheduled languages are covered by the archive programme, with same-week translation to follow as translators are recruited.
- High Courts. Pilot High Courts have adopted the same pipeline for their reportable judgments, with the regional language of the state published within 48 hours and other languages on request.
- Quality. The committee's audit found the machine draft accurate enough to need correction mainly in defined terms, statutory references and names; translators' corrections are fed back to improve the model.
A judgment nobody can read is not a judgment anyone can follow.
A member of the Supreme Court's translation committee, to The Counsel
Why it matters
The courts work in English; most litigants do not. A judgment published only in English reaches a litigant through a lawyer's summary, if at all, and a district court judge in a Hindi-speaking state reads Supreme Court precedent with the same difficulty. Translations within 48 hours mean that a judgment on bail, maintenance or land can be read by the people it affects in the week it is delivered, and that the district judiciary — which the committee identifies as the primary user — can cite it in its own language.
The limits
- Authority. Every translation carries a disclaimer that the English is authoritative. A party cannot rely on the translated text as the court's words, and discrepancies are resolved in favour of the English.
- Coverage. The pipeline covers reportable judgments; the far larger volume of unreported orders is not translated, and the orders that most affect litigants — on interim relief, adjournments and bail — are among them.
- Terminology. Legal terms have no settled equivalents in several languages, and the committee's glossaries are works in progress; the same statute can be rendered differently by different High Courts.
- Translators. The 48-hour promise depends on human reviewers. The registry's sanctioned translator posts are not all filled, and High Court cells rely on retired judicial officers and contract staff.
What is SUVAS?
The Supreme Court Vidhik Anuvaad Software, the court's machine-assisted translation tool launched in 2019. It produces draft translations that registry or High Court translators check before publication; more than 50,000 Supreme Court judgments have been translated into 18 languages.
Where can translated judgments be read?
On the e-SCR portal on the Supreme Court website, searchable by language, and on High Court websites and the National Judicial Data Grid for High Court translations.
Is a translated judgment legally authoritative?
No. The English text is authoritative; translations carry a disclaimer and cannot be relied on as the court's words.
Which High Courts translate their judgments?
Most now have SUVAS cells; Allahabad has published more than a lakh judgments in Hindi, and Patna, Rajasthan, Madhya Pradesh, Bombay, Madras, Kerala, Karnataka, Gujarat, Odisha, Calcutta, Punjab and Haryana and Telangana publish translations in their regional languages.
The Counsel will report the pipeline's extension to the remaining Scheduled languages as the court announces it.
Nikhil Rao
Nikhil covers courts technology, legal AI and the business of law.
This report is for general information only and does not constitute legal advice. Readers should consult a qualified lawyer about their own circumstances. Case citations are to The Counsel Reports (TCR) unless otherwise stated.