Siamsil

non-profit
Activity Feed

AI & ML interests

Neural Machine Translation, Low-Resource Languages, NLP, NLLB-200, Zomi (ctd) Language". This helps researchers and automated indexing tools find your work.

Recent Activity

mangpijasuanΒ  updated a Space 4 days ago
ZomiDatasets/README
mangpijasuanΒ  updated a Space about 1 month ago
ZomiDatasets/README
View all activity

Organization Card

Siamsil πŸŒ„

The digital home for the Zomi language.

Zomi, also called Zolai or Tedim (ISO 639-3 ctd), is spoken by Zomi communities in Myanmar, India, and a growing worldwide diaspora. It has very few digital resources. Siamsil is a non-profit building open tools so Zomi speakers can learn, search, and translate in their own language, and so the language thrives in the age of AI.

What we work on

  • πŸ“– Siamsil Dictionary: an English–Zomi dictionary of about 50,000 entries, being reviewed word by word by native Tedim speakers.
  • πŸ”€ Zomi Standard Form: consistent Zolai spelling, built from community style guides, the Bible, and published Zomi writing.
  • 🌐 Translation: English ⇄ Zomi models fine-tuned from NLLB-200. A model is released only when every source it learned from is properly licensed.
  • πŸŽ™οΈ Speech: consent-based recordings for future Zomi speech recognition and text-to-speech.

Our principles

  • Community first. Native speakers review and approve the language; nothing machine-made is presented as verified.
  • Respect for owners. Every source is credited, and used only as its rights holder allows.
  • Open where we can. Tools and permitted data are shared for the Zomi community and for low-resource language research.

Get involved

  • Tedim dialect speakers: help review the dictionary, a few hundred words at a time, at your own pace.
  • Researchers and developers: we welcome collaboration on low-resource translation and Zomi NLP.
  • Authors, publishers, and churches: share Zomi texts with permission, so your work helps keep the language alive.

Contact us on GitHub: @mangpijasuan

Lungdam mahmah! πŸ™

models 0

None public yet

datasets 0

None public yet