Domain-Aware Error Correction for Citation NER in Medieval Hebrew Responsa

What it is about

The task is citation NER in medieval Hebrew Responsa: each token of a citation gets a role (book name, author, reference number, reference word, and so on, or outside). The authors use an LLM as a tagger driven by a long system prompt, written and refined on the validation set only, that encodes rabbinic abbreviations as lookup anchors and gives priority to reference tokens. Then LLM "judges" flag likely token-level errors, a small gradient-boosting classifier trained on validation data decides which flags to trust, and simple domain rules correct the rest, for example separating quoted lemma text from real citations. Nothing is fine-tuned. On the test set the correction step adds about four F1 points over the prompt baseline, mostly through higher precision, and the few-shot result is 84.4 F1 on test (85.3 on validation). The authors report the system approaching a supervised BEREL-CRF model trained on the domain data.

Ideas

• I1. A flag-then-decide step after the segmenter in tools/hbo: each prefix split is checked by a rule (or by a tiny classifier trained on a held-out slice of MACULA) that flags splits no lexicon entry supports; a flagged split falls back to the unsplit form and the log says why. Measure flag precision and recall against MACULA gold on verses kept out of training.
• I2. A rare-label guard for the tagger: the paper marks rare classes (the adjective class is under 1% of tokens) as rare in the prompt because they are over-predicted. For Mova, compare predicted label rates with the treebank's rates and, when a rare label exceeds its prior, fall back to the default label and record the reason.
• I3. A deterministic table of rabbinic abbreviations (letter pairs with a gershayim, such as the ones for "perek kama" and "amud aleph") as lexicon anchors for the Rabbinic layer, tested on a held-out set of citations.

For Mova

I1 is a check inside tools/hbo, between the segmenter and gparse. I2 is a rule for the output of gparse in tools/hbo, with its measure written up in 13. Biblical Hebrew in Mova: the tools. I3 goes into 11. Mishnaic Hebrew (hbo-x-mishnaic) and is a test for the Rabbinic segments that tools/langseg cuts out.

Topics: Named entity recognition

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.63317/5euewaq3i8b5

@inproceedings{liebeskind2026domain,
  title = {{Domain-Aware Error Correction for Citation NER in Medieval Hebrew Responsa}},
  author = {LIebeskind, Shmuel and Zhitomirsky-Geffet, Maayan and Katzoff, Binyamin and Ben-Gigi, Nati and Schler, Jonathan},
  booktitle = {Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026},
  year = {2026},
  doi = {10.63317/5euewaq3i8b5},
  url = {https://doi.org/10.63317/5euewaq3i8b5}
}

References

16 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Nati Ben-Gigi, Maayan Zhitomirsky-Geffet, Jonathan Schler, and Binyamin Katzoff. 2025. Automatic construction of the citation network from the medieval jewish responsa literature. ACM Journal on Computing and Cultural Heritage, 18(2):1–18. find
2. Monica Berti. 2019. Digital classical philology: Ancient Greek and Latin in the digital revolution, volume 10. Walter de Gruyter GmbH & Co KG. find
3. Marcel Bollmann. 2019. A large-scale comparison of historical text normalization systems. In Proceedings of the 2019 conference of the north American chapter of the association for computational linguistics: Human language technologies, volume 1 (long and short papers), pages 3885– 3898. find our card
4. Giovanni Colavizza, Silvio Peroni, and Matteo Romanello. 2023. The case for the humanities citation index (huci): a citation index by the humanities, for the humanities. International Journal on Digital Libraries, 24(4):191–204. find
5. Isaac Councill, C Lee Giles, and Min-Yen Kan. 2008. Parscit: an open-source crf reference string parsing package. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC’08). find
6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pages 4171–4186. find
7. Maud Ehrmann, Chiara Palladino, and Giuseppe GA Celano. 2016. Diachronic named entity recognition and linking. In Proceedings of the Workshop on Language Technology Resources and Tools for Digital Humanities (LT4DH), pages 1–9. find
8. The COLING 2016 Organizing Committee. find
9. Yaakov HaCohen-Kerner, Nadav Schweitzer, and Yaakov Shoham. 2010. Automatic identification of biblical quotations in hebrew-aramaic documents. In International Conference on Knowledge Discovery and Information Retrieval, volume 2, pages 320–325. SCITEPRESS. find
10. Natallia Kokash, Matteo Romanello, Ernest Suyver, and Giovanni Colavizza. 2024. The brill knowledge graph: A database of bibliographic references and index terms extracted from books in humanities and social sciences. Research Data Journal for the Humanities and Social Sciences, 9(1):1–21. find
11. Guillaume Lample, Miguel Ballesteros, Sandeep Subramanian, Kazuya Kawakami, and Chris Dyer. 2016. Neural architectures for named entity recognition. In Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: human language technologies, pages 260–270. find
12. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. G-eval: Nlg evaluation using gpt-4 with better human alignment. In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 2511–2522. find
13. Michael Piotrowski. 2012. Natural language processing for historical texts. Morgan & Claypool Publishers. find
14. Matteo Romanello. 2013. Creating an annotated corpus for extracting canonical citations from classics-related texts by using active annotation. In International Conference on Intelligent Text Processing and Computational Linguistics, pages 60–76. Springer. find
15. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems, 36:46595– 46623. find
16. Maayan Zhitomirsky-Geffet and Gila Prebor. 2019. Sagebook: Toward a cross-generational social network for the jewish sages’ prosopography. Digital Scholarship in the Humanities, 34(3):676– 695. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2026-liebeskind-domain-aware-error-correction-citation/