Browse project documentation
Tiny Finglish
Choose a setup for converting Latin-script Persian into Persian writing.
Tiny Finglish converts Finglish, Persian typed with Latin letters, into Persian script. It is a transliterator: it does not translate English or rewrite a message into a different register. Its output is a suggestion the writer can review.
Start with the right configuration
Before configure() is called, the module-level transliterate() uses rules without external data. The Eram website uses a hybrid engine: model weights, rules, a frequency table, and a vowel table, with no optional bigram or lexicon. These are different configurations and can produce different text.
| Setup | What you supply | Use it when |
|---|---|---|
| Rules only | Nothing | You want a baseline without asset loading |
| Rules with tables | Decoded frequency and vowel tables | You want dictionary-informed ranking without the neural runtime |
| Hybrid | Model plus both decoded tables | You want the same configuration as the website |
| Model-only generation | Model and hybrid: false; tables optional | You want to compare model candidates; shared pipeline rules still run |
For the verified input salam, emrooz miram shiraz, bare rules return سلم، امروز میرم شیرز; the website configuration returns سلام، امروز میرم شیراز. See the complete Node setup and configuration comparison.
Choose your next step
Start with installation, then use the browser loader or Node loader. Learn how to protect mixed text, read results and candidates, and build a review interface. For typing interfaces, use a worker.
Local processing and downloads
The library runs synchronously on the CPU and makes no network requests. A browser application normally downloads static JavaScript, model, and table assets before converting locally. Downloading those assets is separate from sending input text anywhere. Logging, analytics, persistence, and network policy belong to the application; the library cannot guarantee the privacy of its host page.
Availability and provenance
The v0.1.0 GitHub release contains these guides and the source snapshot. The npm registry returned 404 for tiny-finglish when checked on 9 October 2026; publishing a GitHub release does not publish the npm package. The website uses an archive built from source revision 96b9ff39d9accff72b8d634925833a5d6f337c97. The examples here were verified against that same engine revision. Follow the archive workflow in installation; do not assume an npm registry release exists.
The code is MIT licensed. Retain LICENSE and NOTICE when redistributing. NOTICE records HomoRich data attribution, the optional Lilak lexicon, evaluation-data sources, and deliberate exclusions. Research and contributor documents remain outside public navigation; start with architecture, results, and contributing.