Shoshan — third-party provenance notice
=======================================

The MIT license in LICENSE covers the CODE in this repository. It does not, and
cannot, relicense the model weights or the data they were derived from. Those
carry their own terms, recorded here.

Model weights
-------------
The encoder is fine-tuned from DictaBERT (dicta-il/dictabert) and remains subject
to that model's license.

  https://huggingface.co/dicta-il/dictabert

Training and evaluation data
----------------------------
Derived from the IAHLT Hebrew Universal Dependencies treebank (Knesset and
Wikipedia portions) and from public Hebrew lexicons. The bundled acronym list is
from Wiktionary (CC BY-SA).

Full provenance and terms: docs/DATA_STATEMENT.md
Dataset card:              docs/DATASET_CARD.md

Acknowledgements
----------------
We thank Avner Algom and the IAHLT for the treebank data, and for permission to
release the out-of-domain evaluation sentences.
