Integrating Semantic and Statistical Features for Authorial Clustering of Qumran Scrolls

What it is about

The paper clusters the Hebrew Dead Sea Scrolls to test two long-standing questions: whether the Hodayot hymns (1QHa) split into "Teacher Hymns" and "Community Hymns", and how well sectarian and non-sectarian texts separate. The Hebrew scrolls, with reconstructed text removed, are cut into 978 chunks of 100 words with 15 words of overlap. Each chunk is described by BEREL and AlephBERT embeddings, and by TF–IDF and character-trigram vectors. A graph links chunks whose combined TF–IDF and trigram cosine similarity passes a threshold, and a two-layer graph autoencoder produces embeddings that are clustered with Ward and k-means. The method beats baselines on the scrolls and on a Hebrew Bible validation set, but the classical TF–IDF and trigram features are about as good as BERT. The Teacher Hymns cluster in columns 10–16, which supports a distinct group. The sectarian split is less clear, and the model groups rewritten Bible texts by genre rather than by sect. The code and a dataset are public.

Ideas

• I1. Use the paper's protocol as a test harness for Hebrew clustering in tools/hbo: 100-word chunks, 15-word overlap, and Jaccard against the labelled compositions. Report the score with a deterministic baseline before any embedding.
• I2. Character trigrams plus TF–IDF as the default feature for forms the lexicon does not know. The paper finds these competitive with BERT on fragmentary text, so the SLM can use them for unknown or damaged words in tools/ocrfix before it reaches for a neural model.
• I3. A lexical test of teacher versus community vocabulary: list the words that separate the two hymn groups (the paper's Douglas and Johnson references), and use them as a gold set for a rule-based classifier in world that reads speaker and register.

For Mova

I1 is a test harness for tools/hbo and macula clustering. I2 belongs to tools/ocrfix as the fallback for unknown forms and to The lexicon. I3 is a gold list for world, and a register note for Discourse: narrative, speech and the shape of a text. Before any use, check the licence of the public dataset, which the paper does not state.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.alp-1.2

@inproceedings{lourie2025integrating,
  title = {{Integrating Semantic and Statistical Features for Authorial Clustering of Qumran Scrolls}},
  author = {Lourie, Yonatan and Ben-Dov, Jonathan and Sharan, Roded},
  booktitle = {Proceedings of the Second Workshop on Ancient Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.alp-1.2},
  url = {https://doi.org/10.18653/v1/2025.alp-1.2}
}

References

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

1. Axel Bühler, Gideon Yoffe, Nachum Dershowitz, Eli Piasetzky, Israel Finkelstein, Thomas Römer, and Barak Sober. 2024. Exploring the stylistic uniqueness of the priestly source in genesis and exodus through a statistical/computational lens. Zeitschrift fur die Alttestamentliche Wissenschaft, 136(2):165– 190. Publisher Copyright: © 2024 Walter de Gruyter GmbH. All rights reserved. find
2. Florentino García Martínez. 2010. Beyond the sectarian divide: The" voice of the teacher" as an authority-conferring strategy in some qumran texts. In The Dead Sea Scrolls, pages 227–244. Brill. find
3. Carol A Newsom. 1990. ‘sectually explicit’ literature from qumran. The Hebrew Bible and Its Interpreters, 1:167–87. find
4. John J Collins. 2010. Beyond the Qumran Community: The Sectarian Movement of the Dead Sea Scrolls. Wm. B. Eerdmans Publishing. find
5. Carol A Newsom. 2021. A farewell to the hodayot of the community. Dead Sea Discoveries, 28(1):1–19. find
6. Sanjoy Dasgupta. 2015. A cost function for similarity-based hierarchical clustering. Preprint, arXiv:1510.05043. ↗
7. Renyi Qu, Ruixuan Tu, and Forrest Bao. 2024. Is semantic chunking worth the computational cost? Preprint, arXiv:2410.13070. ↗
8. Devorah Dimant. 2014. History, ideology and Bible interpretation in the Dead Sea Scrolls: Collected studies, volume 90. Mohr Siebeck. find
9. Dirk Roorda. 2019. Text-fabric: handling biblical data with ikea logistics. HIPHIL Novum Journal for Bible and Digital Resources, 5(2):126–135. find our card
10. Peter J. Rousseeuw. 1987. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20:53–65. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2025-lourie-integrating-semantic-statistical-features-authorial/