Blog
Notes from the work
How the corpus, the benchmark and the translator are built, and what we learn along the way. Posts are corrected in place when later work changes the story; each correction is noted.
- 8 October 2026 · 6 min read
How to Type Kashmiri on Your Phone and Computer
Type Kashmiri on Android, iPhone, Mac, Windows and Linux: built-in keyboards, fonts that render ؠ ٲ ۄ correctly, and the Urdu look-alikes to avoid.
Los Angeles County Museum of Art, Public domain, source - 8 October 2026 · 7 min read
Kashmiri Speech Technology: Datasets for TTS and Speech Recognition
Open Kashmiri speech resources as of 2026: Bolbosh TTS, IndicVoices, IndicConformer ASR, OpenSLR and AIKosh sets, with sizes, licences and script issues.
Catlemur, CC BY-SA 4.0, source - 8 October 2026 · 7 min read
Tokenizer Fertility: Why AI Models Charge Kashmiri a ‘Token Tax’
Tokenizer fertility, explained with Kashmiri: why English-centric tokenizers split Perso-Arabic words into many tokens, and what that costs.
Naqshi, Public domain, source - 8 October 2026 · 8 min read
Kashmiri in Unicode: The Letters and Vowel Signs Urdu Doesn't Have
Kashmiri yeh U+0620, the wavy hamza U+065F, the deprecated U+0673 and Kashmiri Devanagari vowels, and why their variants break search and NLP.
Muʻjiz Raqam Khan Qandahari, Public domain, source - 8 October 2026 · 6 min read
Kashmiri's Scripts: Sharada, Perso-Arabic and Devanagari
How Kashmiri came to be written in Sharada, Perso-Arabic, Devanagari and Roman script, and what each means for Unicode, fonts and AI data today.
Ms Sarah Welch (photograph), CC BY-SA 4.0, source - 8 October 2026 · 7 min read
Converting InPage Kashmiri Files to Unicode: A Practical Guide
How to convert Kashmiri InPage (.inp) files to Unicode text, why Urdu converters drop Kashmiri letters and vowel signs, and a checklist to check output.
Kalhana (text); scribe unknown, Public domain, source - 8 October 2026 · 7 min read
Kashmiri OCR: Where Things Stand for Printed Text and Manuscripts
No mature OCR exists for Kashmiri yet. A neutral overview of public datasets, models and their reported limits for Nastaliq print, handwriting and Sharada.
Wellcome Collection, CC BY 4.0, source - 3 April 2026 · 12 min read · updated
A Guide to Kashmiri NLP Resources: Datasets, Models, Papers and Tools
Where Kashmiri language data, models, papers and tools live, what each is good for, and how to start. The live, searchable version is the Resource Hub.
Wellcome Collection, CC BY 4.0, source - 4 March 2026 · 7 min read · updated
Kashmir AI Research: Building an NLP Platform for the Kashmiri Language
Why Kashmiri needs its own research platform, what we built (a corpus, training pipelines and blinded human evaluation), and what has happened since.
Catlemur, CC BY-SA 4.0, source - 4 March 2026 · 6 min read · updated
Kashmiri vs Google Translate: Why AI Still Struggles with Kashmir's Language
Google Translate covers about 240 languages and Kashmiri is still not one of them. Which systems do translate Kashmiri, why quality is low, and what helps.
Muʻjiz Raqam Khan Qandahari, Public domain, source - 4 March 2026 · 7 min read · updated
Low-Resource Language NLP: The Challenge of Building AI for Minority Languages
What makes a language low-resource, the techniques researchers use to work around data scarcity, and what Kashmiri teaches about them.
Ms Sarah Welch (photograph), CC BY-SA 4.0, source - 4 March 2026 · 6 min read · updated
Building the Kashmiri Parallel Corpus: Methodology and Challenges
How two existing public corpora became one 124,102-pair training set: sources, filters, split, licence, and the lessons along the way.
Kalhana (text); scribe unknown, Public domain, source
