Teaching machines Kashmiri
Kashmiri AI exists to give Kashmiri a real place in technology: tools that can read, write, translate and speak the language properly, built with the people who speak it.
01 · The mission
No Kashmiri translation. Yet.
Shabrang Films founder Mir Afnaan Nissar started the project and has been building its foundation, a dataset for the Kashmiri language. It is still in progress. Until it is ready, any Kashmiri on this website will be written by Kashmiri speakers, not guessed by a machine.
ا ب پ ت ٹ ج چ ژ د ڈ ر ڑ س ش ک گ ل م ن ۄ و ہ ؠ ٲ
Kashmiri adds its own letters, such as ٲ, ۄ and ؠ, to the Perso-Arabic script.
02 · Why it isn't live yet
Too little data
Translation tools learn from huge amounts of written and spoken language. For Kashmiri, very little of that exists in digital form, so automatic translations come out unreliable. A studio whose motto is “the actual voice of Kashmir” won't put a machine's guess on its pages.
People in India speak Kashmiri (Census of India, 2011).
Kashmiri is one of India's 22 scheduled languages, and an official language of Jammu & Kashmir since 2020.
Scripts carry it: Perso-Arabic (the official one), Devanagari and the historic Sharada.
Digital resources: little digitised text, recorded speech or translated text for AI to learn from.
03 · How it's being built
Dataset first
AI learns a language from examples. So Kashmiri AI starts with the foundation: real Kashmiri, collected from people and checked by people, then used to train and test models.
Collect
Gather Kashmiri text and recorded speech from people who write and speak it every day, in their own words and dialects.
Check & pair
Correct spelling and meaning, pair sentences with English, Urdu or Hindi, and handle the letters only Kashmiri uses.
Train & test
Combine tools built by the team with open-source models, then test them against real Kashmiri and fix what they get wrong.
04 · Help build it
Speak Kashmiri?
A language tool is only as good as the people it learns from. If you speak, read or write Kashmiri, you can help the dataset grow.