S'gaw Karen · corpus in collection
ကညီကျိာ်
S'gaw Karen · ≈2M speakers · fine-tuning on a community-curated corpus.
The same thought, double the cost.
Provisional counts — corpus in collection. Karen shares the Myanmar Unicode block but not its statistics.
The same block, a different language
S'gaw Karen is written in an extended Myanmar Unicode block — the tokenizer sees the same alphabet as Burmese and treats them as one thing. They are not one thing.
The corpus, still in collection, will name what the base model conflated.
A border, and its libraries
Much of the written record lives in refugee-camp school archives, church hymnals, and diaspora newspapers. Provenance is careful. Consent is per-item and revocable.
This page is set with dashed rules where a claim is not yet community-verified. Do not remove the dashes to make the page look more finished.
We work with open source models. Gemma-4 (and SEA-LION )is being used for Burmese, Karen, and Jinghpaw as well as; Llama-3.1-8B may be used for Rohingya, for byte-level fallback where Hanifi is unrepresented.
Community participation and consent for curation and capture. Provenance recorded and fair compensation provided.
Converting legacy encodings and non-standard scripts into universal digital formats (e.g., Zawgyi to Unicode) is an act of cultural and historic preservation. Working alongside local linguists and community members, we systematically standardize text to be machine-readable while ensuring cultural context, orthographic nuances, and historical integrity remain intact.
Adapting foundational base models to local linguistic realities through targeted, data-efficient techniques calibrated to each language’s structure and context. Training logs and loss curves are published directly to the public archive with complete technical transparency.
Community-annotated benchmarks. Evals and benchmarks designed with local community and experts. Annotators acknowledgements and compensation. Inter-annotator agreement reported alongside every score.
- Q01How to represent Karen orthographic variants across religious and secular writing.
- Q02Whether to include Pwo Karen in a shared corpus or keep it distinct.
The corpus belongs to the communities that made it. Consent is per item and revocable. Revocation is honored in the weights, not just in the archive.
Model weights reflect the terms and governance frameworks chosen with the community.