[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/papers/hbo-2024-staps-large-language-models-biblical-hebrew/)

[home](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [papers](https://md.mova.elamur.ai/free/en/docs/papers/) / [Ancient Hebrew](https://md.mova.elamur.ai/free/en/docs/papers/hbo/) / Large Language Models and Biblical Hebrew: Limitations, pitfalls, opportunities

# Large Language Models and Biblical Hebrew: Limitations, pitfalls, opportunities

**Authors:** [Camil Staps](https://md.mova.elamur.ai/free/en/docs/papers/authors/camil-staps/) · **Year:** 2024 · **Venue:** HIPHIL Novum 9(1), 46–55 **Link:** [tidsskrift.dk/hiphilnovum/article/view/144177](https://tidsskrift.dk/hiphilnovum/article/view/144177) · **Licence:** [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) — cite and paraphrase, never carried into a release

[**0** citations](https://openalex.org/works?filter=cites:W4399353843) openalex-snapshot · 2026-10-14[HIPHIL Novum 9](https://openalex.org/S5406958382)[licence](https://creativecommons.org/licenses/by-sa/4.0)

- [Publisher's page](https://doi.org/10.7146/hn.v9i1.144177)
- [PDF](https://tidsskrift.dk/hiphilnovum/article/download/144177/189562)
- [DOI 10.7146/hn.v9i1.144177](https://doi.org/10.7146/hn.v9i1.144177)
- [OpenAlex](https://openalex.org/W4399353843)

**Authors (1)**

1. [Camil Staps](https://openalex.org/A5031376425) [ORCID](https://orcid.org/0000-0003-1164-6838) · h 2 · 13 citations · 18 works

Citations and author metrics from openalex-snapshot (2026-10-14); authors with the same name are not always told apart.

## What it is about

A critical reply to Elrod (2023; card hbo-2023-elrod-nothing-new-under-sun-study). Staps argues that Elrod's GPT results should be read far less favourably, because of interference on two levels: from modern languages (English, Modern Hebrew) in any Hebrew the model writes, and from translations and commentaries in any answer about a biblical text. Counting tasks are already solved exactly by deterministic tools over tagged data (Text-Fabric/ETCBC), which can also be corrected, so an LLM adds nothing but an interface without accuracy guarantees. GPT's generated "Biblical Hebrew" shows Modern Hebrew spelling and syntax (plene דיבר, a subject-initial asyndetic relative, a missing כי, an odd שמעה). The way forward is task-dependent: small models trained on Biblical Hebrew alone, white-box where the "why" matters (dating, classification), black-box only where accuracy alone counts, measured by benchmarks rather than impressions; the Bible's under 500 000 words limit what any trained model can do.

## Ideas

- I1. An interference check for any Hebrew a teacher model produces: flag Modern Hebrew markers — plene spelling of piel and other forms the Torah writes defective (compare with the Torah dictionary's spelling column), lemmas absent from the Torah lexicon, missing כי/אֲשֶׁר where Biblical syntax requires them. Rule-based, explainable, a gate before any generated Hebrew enters a course.
- I2. A teacher's answer about a verse may come from translations, not the Hebrew; when Mova checks a teacher's reading, compare it with JPS 1917 too, and treat agreement with the translation against the trees as suspect, not as confirmation.
- I3. The paper's division of labour is Mova's: exact counts from the trees, models only where no exact method exists, white-box first. Cite it in the tools chapter as outside support.

## For Mova

I1 is a small checker beside `macula dict` (spelling and lexicon from the Torah) for the Torah course and any generated Hebrew; its rules belong in [12. Biblical, Mishnaic and Modern Hebrew compared](https://md.mova.elamur.ai/free/en/docs/languages/hbo/12-biblical-mishnaic-modern/) (Modern vs Biblical differences). I2 is a rule for teacher checks in `tools/macula/src/prag.rs`'s agreement measure with JPS. I3 is a paragraph in `13-mova-tools.md`. The paper is CC BY-SA: paraphrased only.

## Cite

The paper itself, not this card. BibTeX from the library's record of the work. [DOI 10.7146/hn.v9i1.144177](https://doi.org/10.7146/hn.v9i1.144177)

```
@inproceedings{staps2024large,
  title = {{Large Language Models and Biblical Hebrew: Limitations, pitfalls, opportunities}},
  author = {Staps, Camil},
  booktitle = {HIPHIL Novum 9},
  year = {2024},
  doi = {10.7146/hn.v9i1.144177},
  url = {https://doi.org/10.7146/hn.v9i1.144177}
}
```

- [← previous D-Nikud: Enhancing Hebrew Diacritization with LSTM and Pretrained Models](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2024-rosenthal-d-nikud-enhancing-hebrew-diacritization/)
- [next → A Phase-Theoretic Approach to Serial Verb Constructions in Semitic](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2025-altamimi-phase-theoretic-approach-serial-verb/)
- [↑ back to Ancient Hebrew](https://md.mova.elamur.ai/free/en/docs/papers/hbo/)

## References

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

1. Amgoud, Leila. 2023. ‘Explaining black-box classifiers: Properties and functions’. International Journal of Approximate Reasoning 155:40–65. https://doi.org/10.1016/j.ijar.2023.01.004 [↗](https://doi.org/10.1016/j.ijar.2023.01.004)
2. Assael, Yannis, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag & Nando de Freitas. 2022. ‘Restoring and attributing ancient texts using deep neural networks’. Nature 603(7900):280–283. https://doi.org/10.1038/s41586-022-04448-z [↗](https://doi.org/10.1038/s41586-022-04448-z)
3. Bender, Emily M., Timnit Gebru, Angelina McMillan-Major & Shmargaret Shmitchell. 2021. ‘On the dangers of stochastic parrots: Can language models be too big? ’. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–623. ACM. https://doi.org/10.1145/3442188.3445922 [↗](https://doi.org/10.1145/3442188.3445922)
4. Bird, Steven & Haejoong Lee. 2007. ‘Graphical query for linguistic treebanks’. In Proceedings, PACLING 2007 – 10 th Conference of the Pacific Association for Computational Linguistics. Melbourne. [http://hdl.handle.net/11343/34835](http://hdl.handle.net/11343/34835) [↗](http://hdl.handle.net/11343/34835)
5. Elrod, A. G. 2023. ‘Nothing new under the sun? The study of Biblical Hebrew in the era of generative pre-trained AI’. HIPHIL Novum 8(2):1–32. https://doi.org/10.7146/hn.v8i2.143114 [↗](https://doi.org/10.7146/hn.v8i2.143114) [our card](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2023-elrod-nothing-new-under-sun-study/)
6. Elrod, A. G. 2024. ‘Uncovering theological and ethical biases in LLMs: An integrated hermeneutical approach employing texts from the Hebrew Bible’. HIPHIL Novum 9(1):2–45. https://doi.org/10.7146/hn.v9i1.143407 [↗](https://doi.org/10.7146/hn.v9i1.143407)
7. Naaijer, Martijn, Constantijn Sikkel, Mathias Coeckelbergs, Jisk Attema & Willem Th. van Peursen. 2023. ‘A Transformer-based parser for Syriac morphology’. In Proceedings of the Ancient Language Processing Workshop associated with RANLP-2023, 23–29. https://aclanthology.org/2023.alp-1.3 [↗](https://aclanthology.org/2023.alp-1.3/) [our card](https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2023-naaijer-transformer-based-parser-syriac-morphology/)
8. OpenAI. 2023. ‘GPT-4 technical report’. https://doi.org/10.48550/arXiv.2303.08774 Rillig, Matthias C., Marlene Ågerstrand, Mohan Bi, Kenneth A. Gould & Uli Sauerland. ‘Risks and benefits of Large Language Models for the environment’. Environmental Science & Technology 57(9):3464– 3466. https://doi.org/10.1021/acs.est.3c01106 [↗](https://doi.org/10.48550/arxiv.2303.08774)
9. This issue is not specific to LLMs, of course. For example, the Tiberias stylistic classifier for the Hebrew Bible is likewise closed source ([https://tiberias.dicta.org.il/](https://tiberias.dicta.org.il/), retrieved February 9, 2024). [↗](https://tiberias.dicta.org.il/)
10. Roorda, Dirk. 2017–2022. Annotation/text-fabric. Zenodo. https://doi.org/10.5281/zenodo.592193 Roorda, Dirk, Christiaan Erwich, Cody Kingham & SeHoon Park. 2017–2023. ETCBC/bhsa. Zenodo. https://doi.org/10.5281/zenodo.10049740 [↗](https://doi.org/10.5281/zenodo.592193)
11. Rudin, Cynthia. 2019. ‘Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead’. Nature Machine Intelligence 1(5):206–215. https://doi.org/10.1038/s42256-019-0048-x [↗](https://doi.org/10.1038/s42256-019-0048-x)
12. Strubell, Emma, Ananya Ganesh & Andrew McCallum. 2019. ‘Energy and policy considerations for deep learning in NLP’. In Proceedings of the 57 th Annual Meeting of the Association for Computational Linguistics, 3645–3650. Florence: Association for Computational Linguistics. https://doi.org/10.18653/v1/P19-1355 [↗](https://doi.org/10.18653/v1/p19-1355)
13. Tunstall, Lewis, Leandro von Werra & Thomas Wolf. 2022. Natural Language Processing with Transformers: Building language applications with Hugging Face. Beijing: O’Reilly. [find](https://scholar.google.com/scholar?q=Tunstall%2C+Lewis%2C+Leandro+von+Werra+%26+Thomas+Wolf%2E+2022%2E+Natural+Language+Processing+with+Transformers%3A+Building+language+applications+with+Hugging+Face%2E+Beijing%3A+O%E2%80%99Reilly%2E)
14. Van de Bijl, Etienne P., Cody Kingham, Wido van Peursen & Sandjai Bhulai. 2019. ‘A probabilistic approach to syntactic variation in Biblical Hebrew’. https://doi.org/10.5281/zenodo.2546802 [↗](https://doi.org/10.5281/zenodo.2546802) [our card](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2019-bijl-probabilistic-approach-syntactic-variation-biblical/)
15. Van der Schans, Yanniek, David Ruhe, Wido van Peursen & Sandjai Bhulai. 2020. ‘Clustering biblical texts using recurrent neural networks’. https://doi.org/10.5281/zenodo.4003509 [↗](https://doi.org/10.5281/zenodo.4003509) [our card](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2020-schans-clustering-biblical-texts-using-recurrent/)
16. Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser & Illia Polosukhin. 2017. ‘Attention is all you need’. Advances in Neural Information Processing Systems 30 (NIPS 2017). https://doi.org/10.48550/arXiv.1706.03762 [↗](https://doi.org/10.48550/arxiv.1706.03762)
17. Wilson-Wright, Aren. 2023. ‘COHeN’. [https://huggingface.co/gngpostalsrvc/COHeN](https://huggingface.co/gngpostalsrvc/COHeN), version 82ff154, retrieved March 18, 2024. [↗](https://huggingface.co/gngpostalsrvc/COHeN)
18. Young, Ian & Robert Rezetko. 2008. Linguistic dating of biblical texts. London: Equinox. [find](https://scholar.google.com/scholar?q=Young%2C+Ian+%26+Robert+Rezetko%2E+2008%2E+Linguistic+dating+of+biblical+texts%2E+London%3A+Equinox%2E)