# Chavruti derived data: Talmud Bavli (Wikisource edition)

Word-level data that Chavruti (https://chavruti.ai) derived from the Talmud Bavli (Wikisource edition), shared under the license of the source
text: **CC BY-SA 4.0** (https://creativecommons.org/licenses/by-sa/4.0/).

## Source

Talmud Bavli, Wikisource edition (he.wikisource.org/wiki/תלמוד_בבלי, by the volunteers of Hebrew Wikisource), via Sefaria (https://www.sefaria.org). License: CC BY-SA 4.0 (Sefaria: CC-BY-SA; Wikisource: Creative Commons Attribution-Share Alike 4.0). Chavruti's segmentation, word tokens and lemma links on this text are shared under the same license, published at https://chavruti.ai/data.

Source text: Hebrew Wikisource, תלמוד בבלי (https://he.wikisource.org/wiki/תלמוד_בבלי), as published by Sefaria
(https://www.sefaria.org, version "Wikisource Talmud Bavli", license CC-BY-SA). Wikisource states its license as
"Creative Commons Attribution-Share Alike 4.0". The text is unvocalized.

## What is here

- `segments.jsonl.gz`: one JSON object per line: `ref` ("Berakhot 2a:1"), `tractate`, `amud` ("2a"), `position`
  (1-based within the amud), `text` (the segment as loaded: HTML removed, whitespace collapsed) and `tokens`, a list of
  `{"i", "raw", "form", "lemma"}`: word order, the word as written in the segment, our normalized form key (the
  unpointed spelling; abbreviations keep their gershayim), and a lemma id or null; `"how": "citation"` when the
  lemma is this occurrence's own (Jastrow cites this passage for it) rather than its form's.
- `forms.jsonl.gz`: one line per distinct form: `form`, `consonants`, `lemma`, `tokens` (occurrences here), and, for
  a link made by matching Jastrow, `how` (exact, affix, dominant) and `alternates` (other entries the letters fit).
- `manifest.json`: counts and when this was made.

Lemma ids are namespaced: `oshb:` (Open Scriptures Hebrew Bible lexicon, CC BY 4.0), `wikt:` (English Wiktionary,
CC BY-SA 4.0), `jastrow:` (Jastrow's Dictionary of the Targumim, the Talmud Babli and Yerushalmi). Links without
`how` were made by spelling agreement: every vocalized form with the same consonants in Chavruti's other texts has
that one lemma, and the spelling has 4 letters or more. The rest match Jastrow's entries by their letters, prefixes
and suffixes, how often Jastrow cites each, and the passages Jastrow cites; checked by hand, about 8 to 9 in 10 are
right depending on the method. Treat the links as candidates, not as a morphological analysis.

## License

CC BY-SA 4.0. You may share and adapt this data, for any purpose, including commercially, if you give appropriate
credit (the attribution above, and "derived data by Chavruti, chavruti.ai"), link to the license, say if you changed
it, and share what you build from it under the same license.

To be published at https://chavruti.ai/data. Generated 2026-11-09T16:00:29Z (ingest version 3).