TigreGotico/hebrew_diacritized_text

Hebrew Diacritized Text

~5.3 million Hebrew sentences fully annotated with nikud (vowel diacritics), sourced from the phonikud project's knesset_nikud_v6 corpus (Knesset parliamentary transcripts diacritized with Dicta, then iteratively corrected through six refinement passes).

Intended for training Hebrew diacritization models analogous to rawi for Arabic.

Format

Plain text, one sentence per line, UTF-8. Files: train.txt, validation.txt, test.txt.

מָה עִם אֵ֫לֶּה שֶׁ|חַיִּים מִ|קִּצְבָּאוֹת שֶׁל אֲנָשִׁים עִם מֻו֯גְבָּלוּיוּת?

Splits

Split	Sentences	Share
train	4,529,821	85 %
validation	266,465	5 %
test	532,925	10 %

Splits are stratified by sentence length and unique character count (20 × 10 bins) so all three partitions share the same length and character-diversity distribution.

Source & License

Derived from thewh1teagle/phonikud-data (MIT) and IsraParlTweet (CC-BY-4.0). Released under CC-BY-4.0.