Sefaria pointed Hebrew corpus as used by rababa

Purpose

Pointed (nikud + dagesh + sin) Hebrew text used to train the rababa Hebrew diacritization model.

Source

All text fetched from the Sefaria API (developers.sefaria.org). Sefaria is a non-profit organization that makes Jewish texts freely available.

The fetch script lives at scripts/fetch_sefaria_corpus.py.

Books included

• Tanakh (39 books): Torah + Nevi’im + Ketuvim — fully pointed Biblical Hebrew.

• Mishnah (63 tractates across 6 sedarim): rabbinic legal text.

• Siddurim (Ashkenaz, Sefard, Edot HaMizrach): prayer books.

Splits

This is a Biblical + Rabbinic Hebrew corpus, not Modern Hebrew. Diurnal usage differs from modern (e.g., grammar, vocabulary). For Modern Hebrew, see the distillation-augmented corpus in rababa-modern-hebrew-distilled (TBD).

License

Sefaria texts are public domain or various open licenses (CC-BY, CC-0). See Sefaria’s licensing page for per-text details. This compiled dataset is provided for unencumbered ML training access.

Stats

Total lines: 18,628

Split	Lines
train	14,902
val	1,862
test	1,864

Layout

sefaria_train/train.txt
sefaria_val/val.txt
sefaria_test/test.txt

Matches the layout of interscript/rababa-tashkeela.