A New Hebrew Universal Dependency Treebank: The First Treebank of Post-Rabbinic Historical Hebrew

What it is about

The authors release the first UD treebank of post-Rabbinic historical Hebrew: about 11,800 words from six works (Igeret orhot olam, Sefer ha-hinukh, Rashi's Pentateuch commentary, Shulhan arukh, Sefer Maharil and a miscellany), written between the tenth and nineteenth centuries. Four linguists annotated each sentence and then reviewed every annotation together. The paper argues that this layer differs from Modern Hebrew in ways that break existing parsers: dislocated topics with no punctuation (37 sentences, 11.5%, against about 1% or less in the modern corpora), causal clauses with bare she- that look like relatives (more than 15 sentences), participles that are verbs in form but subjects or genitive heads in function, and verbs of different tenses coordinated inside one clause. A Modern Hebrew parser (DictaBERT) reads the dislocated subject as the subject of the main clause. Because the corpus is below 20K words, the authors do not split it and recommend 10-fold cross-validation. They add subtypes of existing UD labels (such as a participle subtype) so that the corpus stays valid for cross-lingual comparison.

Ideas

• I1. A dislocated-topic rule for post-biblical text: a noun phrase at the start of a sentence, with no punctuation before the main verb, and a possessive pronoun later in the clause that points back to it ("his father" for "a small boy ... his father") is a topic, not the subject. Write it as a rule and test it on the 37 cases the paper lists with their sentence IDs.
• I2. Bare she- causal clauses: when she- follows a clause boundary with no head noun, a rule can use the rest of the sentence (a legal or ritual consequent) to choose causal over relative. For Mova this is a cause relation in the world, with a flag that says the reading was chosen by a rule, not by a parser.
• I3. Evaluation protocol: with a corpus under 20K words, use 10-fold cross-validation, not a fixed split, and report the mean. Every small Mova test set on rare constructions should follow the same rule, so that no single split decides a decision.

For Mova

I1 goes into Syntax: the clause and the sentence (topic and subject) as a rule with a test list; the same contrast is a useful note for 7. Syntax. I2 goes into Discourse: narrative, speech and the shape of a text as a causal-clause rule, and into the cause relation of world. I3 is the evaluation rule for any test on treebanks/he/postrab, which is used only as a test (never as training data for a release).

Topics: Treebanks

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.5281/zenodo.17216880

@inproceedings{tal2025new,
  title = {{A New Hebrew Universal Dependency Treebank: The First Treebank of Post-Rabbinic Historical Hebrew}},
  author = {Tal, Rachel and Fuchs, Shlomit and Albeck, Orly and Brauner, Elisheva and Lindenbaum, Yitzchak and Meiri, Ephraim and Shmidman, Avi},
  booktitle = {Proceedings of the 23rd International Workshop on Treebanks and Linguistic Theories (TLT, SyntaxFest 2025)},
  year = {2025},
  doi = {10.5281/zenodo.17216880},
  url = {https://doi.org/10.5281/zenodo.17216880}
}

References

9 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Meni Adler, Yael Dahan Netzer, Yoav Goldberg, David Gabay, and Michael Elhadad. 2008. Tagging a Hebrew corpus: the case of participles. In LREC. find our card
2. Daniel G Swanson, Bryce D Bussert, and Francis Tyers. 2024. Producing a parallel universal dependencies treebank of Ancient Hebrew and Ancient Greek via cross-lingual projection. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 13074–13078. find our card
3. Zeev Ben-Hayyim. 1985. The problems of the unity of the Hebrew language throughout its history and 12 https://github.com/ERC-Midrash/UD_ ↗
4. Hebrew-PostRab David Tènè. 1985. Historical identity and unity of Hebrew and the division of its history into periods. In Moshe Bar-Asher, editor, Language Studies, volume 1. Magnes. find
5. Universal Dependencies. Universal dependencies -nested coordination. Accessed: 2025-04-07. find
6. Universal Dependencies Contributors. 2024. UD for Hebrew. Accessed: 2025-06-12. find
7. Yoad Winter, Alon Altman, K. Sima’an, Alon Itai, and Noam Nativ. 2001. Building a tree-bank of modern hebrew text. find
8. Amir Zeldes, Nick Howell, Noam Ordan, and Yifat Ben Moshe. 2022. A second wave of UD Hebrew treebanking and cross-domain parsing. find our card
9. Tamar Zewi and Yael Reshef. 2009. The active participle and temporal expression in Hebrew. Leshonenu, 71(3–4):315–344. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/ud-2025-tal-new-hebrew-universal-dependency-treebank/