Modeling the Dalet Clitic in Historical Hebrew Texts: A New Prefix-Segmented BERT Model and Stylistic Analysis

What it is about

The Aramaic proclitic dalet in historical Hebrew can mark a subordinate clause or a possessive preposition, and the two uses look the same on the page; no annotated data existed. The authors pretrain a BERT model on 810 million words of Rabbinic Hebrew in which every prefix is a separate token, using a prefix-segmentation classifier they built for this. For each dalet they read the masked-LM prediction at that position and ask whether the Hebrew forms for the subordinator or for the preposition appear in the top k predictions (k = 500 in the best setting); if neither appears, the item is "unknown". On eight rabbinic works annotated by hand, the masked setting reaches an average F1 above 0.89. Applied to more than 300 million words, the method shows geographic and diachronic trends and stylistic clusters. Recall for the minority preposition class stays at about 71.8 when the clitic is left visible, which the authors read as a bias toward the majority class. The annotated dataset and the model are released.

Ideas

• I1. A rule module for the dalet clitic: the paper cites fixed constructions in which dalet is a subordinator (such as "kol" followed by a time or place word and dalet). Write them as a small deterministic table for Aramaic and Rabbinic tokens, return "unknown" when no rule fires, and test it on the released eight-work dataset.
• I2. Keep "unknown" as a legitimate answer, as the paper does, instead of forcing a guess. A Mova rule that cannot decide should say so and name what it tried.
• I3. Prefix as its own token: Mova's tools/hbo segmenter already splits prefixes, so the dalet can be a separate word with its own label for gparse; the released dataset gives a gold test for that label.

For Mova

I1 goes into Particles and syntax of Biblical Aramaic for the Aramaic particle and into languages/tmr/ for Babylonian Talmudic usage. I2 is a design rule for these tables. I3 is a test for tools/hbo and a note in Pronouns, particles and prepositions. The licence of the released dataset must be checked before it is used as a measure. The masked-LM probe itself is not a Mova target.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.nlp4dh-1.12

@inproceedings{tal2026modeling,
  title = {{Modeling the Dalet Clitic in Historical Hebrew Texts: A New Prefix-Segmented BERT Model and Stylistic Analysis}},
  author = {Tal, Rachel and Shmidman, Cheyn Shmuel and Shmidman, Avi},
  booktitle = {Proceedings of the 6th International Conference on Natural Language Processing for the Digital Humanities},
  year = {2026},
  doi = {10.18653/v1/2026.nlp4dh-1.12},
  url = {https://doi.org/10.18653/v1/2026.nlp4dh-1.12}
}

References

14 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Moshe Asis. 2010. A Treasury of Jerusalemite Expressions. The Jewish Theological Seminary of America, New York and Jerusalem. Hebrew; published in 2010 (Hebrew year 5770). find
2. Yitzhak Baer. 1958. Toledot ha-Yehudim bi-Sefarad ha-Notzrit. Am Oved, Tel Aviv. 2nd rev. and expanded ed.; Hebrew. find
3. Elitzur A. Bar-Asher Siegal. 2016. Introduction to the Grammar of Jewish Babylonian Aramaic, 2 edition. Ugarit-Verlag, Münster. References: §§ 10.1, 4.3. find
4. Haim Beinart. 1988. Jewish–converso relations between spain and italy. In Haim Beinart, editor, Jews in Italy: Studies Dedicated to the Memory of U. Cassuto on the 100th Anniversary of His Birth, pages 275–288. Magnes Press, Jerusalem. find
5. Ram Ben-Shalom. 2008. The Jews of Provence and Languedoc. Magnes Press, Jerusalem. find
6. Esther Benbassa and Aron Rodrigue. 2001. The Jews of the Balkans: The Judeo-Spanish Community, 15th–20th Centuries. The Zalman Shazar Center for Jewish History, Jerusalem. Hebrew translation. Originally published in French. find
7. Yochanan Breuer. 2007. The babylonian aramaic in tractate karetot: According to ms oxford. Aramaic Studies, 5(1):1–45. ↗
8. Yosef Kaplan. 2021. Jews on the move: Early modern jewish migration. The Posen Library of Jewish Culture and Civilization. find
9. Ivan G. Marcus. 2010. Ashkenaz. YIVO Encyclopedia of Jews in Eastern Europe, accessed 2026-03-12. find
10. Timo Schick and Hinrich Schütze. 2021. Exploiting cloze-questions for few-shot text classification and natural language inference. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 255–269, Online. Association for Computational Linguistics. ↗
11. Avi Shmidman, Joshua Guedalia, Shaltiel Shmidman, Cheyn Shmuel Shmidman, Eli Handel, and Moshe Koppel. 2022. Introducing berel: Bert embeddings for rabbinic-encoded language. Preprint, arXiv:2208.01875. ↗ our card
12. Avi Shmidman, Ometz Shmidman, Hillel Gershuni, and Moshe Koppel. 2024a. MsBERT: A new model for the reconstruction of lacunae in Hebrew manuscripts. In Proceedings of the 1st Workshop on Machine Learning for Ancient Languages (ML4AL 2024), pages 13–18, Hybrid in Bangkok, Thailand and online. Association for Computational Linguistics. ↗ our card
13. Shaltiel Shmidman, Avi Shmidman, and Moshe Koppel. 2023. Dictabert: A state-of-the-art bert suite for modern hebrew. Preprint, arXiv:2308.16687. ↗ our card
14. Shaltiel Shmidman, Avi Shmidman, Moshe Koppel, and Reut Tsarfaty. 2024b. MRL parsing without tears: The case of Hebrew. In Findings of the Association for Computational Linguistics ACL 2024, pages 4537–4550, Bangkok, Thailand and virtual meeting. Association for Computational Linguistics. A ↗ our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2026-tal-modeling-dalet-clitic-historical-hebrew/