CACCHT

Creating Annotated Corpora of Classical Hebrew Texts

Ancient Semitic
texts, published
as data.

CACCHT prepares linguistically annotated editions of ancient Semitic texts and releases them openly, so that a question about grammar, syntax or textual variation can be answered by running code over a whole corpus.

CACCHT is an initiative of Robert Rezetko and Martijn Naaijer (PI).
Get in touch: martijn.naaijer@uzh.ch

Seven corporaFive scriptsOne annotation system

Browse the corpora The GitHub organization
The CACCHT roundel: aleph, alpha, olaph and the Ugaritic alpa in four quadrants
Hebrew · Greek · Syriac · Ugaritic

Corpora

Seven datasets, free to use

Every dataset is a Text-Fabric dataset: text and annotations stored as a graph of nodes and features, read straight into Python. Some carry word-level annotation only; others add phrase and clause structure.

אReleased

The CACCHT Dead Sea Scrolls

Biblical and non-biblical scrolls, transcribed and annotated word by word.

Source
Transcriptions provided by Martin Abegg

RepositoryDOICC BY-NC 4.0

אEarly release

The Aleppo Codex

The Hebrew Bible according to the Aleppo Codex, segmented and annotated in BHSA style, with every word mapped to its BHSA word and every verse marked for whether it survives in the codex.

Source
מקרא על פי המסורה, Hebrew Wikisource

RepositoryCC BY-SA 4.0

ܐIn progress

The ETCBC Syriac Corpus

Peshitta books, the Syrohexapla Psalms, and post-biblical prose including Ephrem and the Book of the Laws of the Countries.

Source
The ETCBC database of Syriac literature

RepositoryDOICC BY-NC 4.0

ࠀIn progress

The CACCHT Samaritan Pentateuch

The Samaritan Pentateuch with word-level annotations in BHSA style, plus phrase-atom and clause-atom boundaries.

Source
Samaritanus project, Halle-Wittenberg (ed. Stefan Schorch): MS Dublin Chester Beatty 751 and MS Garizim 1

RepositoryDOICC BY-NC 4.0

אEarly release

The ETCBC Targum Corpus

Pseudo-Jonathan, the Fragment Targums, the Cairo Genizah fragments and Neofiti — with scribal variants kept as first-order data rather than flattened away.

Pipeline
SQL to XML to Text-Fabric, included in the repository

RepositoryDOICC BY-NC 4.0

𐎀In progress

The Copenhagen Ugaritic Corpus

278 tablets of Die keilalphabetischen Texte aus Ugarit, annotated down to tablet, column, line and side, with certainty marked on individual signs and alternative readings recorded.

With
Tania Notarius (University of the Free State), Alex Sosnovshchenko and Kseniia Yerofeieva

RepositoryDOICC BY-NC 4.0

ΑIn development

The Septuagint

A Text-Fabric edition of the Greek text, under construction at CenterBLC. The repository holds the conversion pipeline and the dataset as it grows.

Status
No release yet — follow the repository

Repository

Most CACCHT datasets are licensed CC BY-NC 4.0: free for research, teaching and non-commercial applications, while commercial use needs permission. Read the data-use statement →

Tools

Reading the witnesses side by side

Browser tools built on annotated data. Each one sets parallel witnesses next to each other, chapter by chapter and verse by verse, and is free to use. All of them are work in progress, so expect the data and the interfaces to keep changing.

Tool MT–SP Parallels Read the Masoretic Text and the Samaritan Pentateuch side by side, verse by verse, built on the BHSA and the CACCHT SP data. Tool The Hebrew Bible: MT and the Dead Sea Scrolls Every verse of the Masoretic Text next to whatever the Judaean Desert manuscripts preserve of it, across 262 scrolls. Reconstructed letters are marked as such, and the chapter index shows at a glance which chapters are attested at all. Tool The Short Edition of Jeremiah at Qumran Jeremiah in three traditions: the Masoretic Text, the Old Greek and the six Jeremiah scrolls from the Judaean Desert. It shows where 4QJerᵇ and 4QJerᵈ follow the shorter Greek edition, down to whether the missing verses would physically fit the fragment. Tool Textkritische Synopse zum Jeremiabuch A synopsis by Hermann-Josef Stipp: the Hebrew of Jeremiah beside a Hebrew retroversion of the Alexandrian Greek, for all 52 chapters plus 2 Kings 25. Hover a word for its lexeme and grammatical parsing, and ketiv-qere readings are flagged.

Characteristic language per book

Tools that look at the vocabulary of the Hebrew Bible as a whole, showing what sets each book apart from the rest.

Tool Characteristic Language in the Hebrew Bible The words and phrases that are unique to a book, preferred by it or rarely shared with others, drawn from the BHSA. Filter by book, chapter and word category, and for Jeremiah and 2 Kings 25 include the Masoretic–Septuagint differences from Stipp's synopsis.

Method

One annotation system, adapted per language

We follow the BHSA

The Biblia Hebraica Stuttgartensia Amstelodamensis — the annotated Masoretic Text maintained by the ETCBC — sets the conventions. CACCHT follows them and adapts them where a language or a text demands it, so that features keep the same meaning across corpora and a query written for one dataset largely survives the move to another.

g_conslexglossspvsvtpsgnnuvostlstrailer

Word-level features in the Syriac corpus. Each dataset documents its own set.

Read it with Text-Fabric

Nothing needs downloading by hand. Point Text-Fabric at a dataset and it fetches the data, then gives you nodes, features and a search language over them — in a Jupyter notebook, or in the browser it ships with.

The same datasets can be read with Context-Fabric, an API-compatible engine that loads them with a far smaller memory footprint and exposes them to AI agents over MCP.

from tf.app import use
A = use('dt-ucph/sp')
tf dt-ucph/sp

The Samaritan Pentateuch, in a notebook and in the Text-Fabric browser.

Open by default

Every dataset lives in a public repository, and each release is archived with a DOI at Zenodo. All of them are free to use for research and education. Licenses differ per corpus, so check the card above, read the data-use statement, and cite the release you actually used.

Publications

How the annotations are made, written up

  1. 2026

    Identifying Phrase Boundaries in the Samaritan Pentateuch with Machine Learning

    Cantanhêde, S. d. O., Naaijer, M., Højgaard, C. C., & Glanz, O. — Religions 17(2), 192.

    10.3390/rel17020192

  2. 2024

    Text-Fabric Dataset of the Samaritan Pentateuch

    Naaijer, M., Højgaard, C. C., Schorch, S., & Ehrensvärd, M. — Research Data Journal for the Humanities and Social Sciences 9(1), 1–13.

    10.1163/24523666-bja10051

  3. 2023

    A Transformer-based parser for Syriac morphology

    Naaijer, M., Sikkel, C., Coeckelbergs, M., Attema, J., & Van Peursen, W. Th. — Proceedings of the Ancient Language Processing Workshop, Varna, 23–29.

    ACL Anthology

Team

A collaboration of six

CACCHT works together with specialists in each field to build the datasets.

  • Martijn NaaijerUniversity of Zurich
  • Willem van PeursenVrije Universiteit Amsterdam
  • Oliver GlanzAndrews University
  • Christian Canu HøjgaardFjellhaug International University College
  • Martin EhrensvärdUniversity of Copenhagen
  • Robert RezetkoUniversity of Arizona and University of Copenhagen

With Stefan Schorch and the Samaritanus project, Martin Abegg, Tania Notarius, Alex Sosnovshchenko, Kseniia Yerofeieva, and colleagues at the ETCBC.

Use the data. Cite the release.

The datasets are free for research and education. Each repository states how to cite it; the data-use statement explains what the license allows.

Open the GitHub organization