Elamur Research⌕
☰
home / papers / 2026 / Mining the Pre-1948 Palestinian Press: Unsupervised Keyphrase Extraction and Temporal Discourse Analysis from Five Historical Arabic Newspapers

Mining the Pre-1948 Palestinian Press: Unsupervised Keyphrase Extraction and Temporal Discourse Analysis from Five Historical Arabic Newspapers

Paper · 2026 · Basel Barakat and Nizam Barakat

Summary

The authors tackle the lack of computational access to the Arabic-language Palestinian press from before 1948. Most of this material survives only as scanned pages in archives. They build a four-stage pipeline covering acquisition, OCR, keyphrase extraction and visualization, and apply it to five newspapers: Lisān al-ʿArab, Al-Bushrā, Al-Karmil, Al-Difāʿ and Filasṭīn. Page images are transcribed with the GCV-OCR API, and words with low recognition confidence are dropped. The text is then split into dated document units. KeyBERT, using AraBERT as its encoder, ranks candidate n-grams by how close their embeddings are to the embedding of the whole document, so no labelled data is needed. A Python toolkit then draws heatmaps, trend lines and bar charts that render Arabic script correctly. The extracted keyphrases differ by newspaper in ways that match each paper's editorial line, from pan-Arab nationalism to religious and communal affairs. Over time, keyphrase trends follow major events such as the 1936–1939 Arab Revolt and the growing talk of sovereignty near 1948. The work matters because it turns a neglected historical archive into material that NLP tools can study.

Key points

  • The corpus is made up of five subcorpora spanning the final decades of Ottoman rule up to the close of the British Mandate:
    • Al-Karmil (Haifa, 1908–1944)
    • Filasṭīn (Jaffa, 1911–1948)
    • Al-Difāʿ (Jaffa, 1934–1948)
    • Lisān al-ʿArab
    • Al-Bushrā
  • OCR uses the GCV-OCR DOCUMENT_TEXT_DETECTION mode, which suits dense multi-column layouts and returns a confidence score for every word. Post-processing normalizes encoding, filters out low-confidence words and produces one CSV per newspaper.
  • Extraction is KeyBERT with an AraBERT backbone, written up formally as a top-k argmax over the cosine similarity between candidate and document embeddings.
  • The visualization toolkit handles right-to-left display through character reshaping plus BiDi reordering. The code and data format specs are released as supplementary material.
  • Acknowledged limitations:
    • OCR is noisy, especially for pre-1920 pages.
    • AraBERT was pre-trained on Modern Standard Arabic (MSA), which differs from early-20th-century Arabic, so there is a domain shift.
    • Newspaper-name collocations from the repeated masthead show up among the top keyphrases.

Interesting ideas

  • I1. Using the recognizer's own per-word confidence to discard tokens before analysis keeps noisy input from polluting later stages without any manual cleaning.
  • I2. Text that repeats on every page, like a masthead, can float to the top of a document-similarity ranking and pose as content; the authors flag this as an artifact to filter.
  • I3. Splitting pages into date-stamped units is what makes per-period comparison and event alignment possible in the first place.
  • I4. The paper keeps "keyphrase" (multi-word n-gram unit) and "keyword" (token-level counts) as separate terms, which avoids confusing two different granularities in evaluation.
  • I5. Salience is scored by how close a candidate is to a representation of the whole document, needing no labels.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): silver data could likewise drop arcs or tokens whose witness-parser agreement or confidence falls below a threshold, rather than training on everything.
  • I2 (learn): recurring boilerplate (titles, headers, formulaic openings) could serve as a negative control, so that any "salient" output dominated by it signals a bug.
  • I3 (world): attaching a document date to each text's world would let event order and state changes be anchored to external time across a collection, not just within one story.
  • I4 (tokens): stating explicitly whether a statistic counts lemmas, tokens or multiword expressions would keep tagger and lemma reports comparable.
  • I5 (back): a structural analogue could rank entities and frames by how central they are in the world graph, choosing what the compressor or summary generator keeps without falling back on bag-of-words statistics.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Goldsmiths University of LondonLondon · United Kingdom4 works · 381 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksGeneral Social Sciences · 227 worksLiterature and Literary Theory · 519 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksDiscourse analysis · 466 works

Cite

Cite as Basel Barakat and Nizam Barakat. 2026. Mining the Pre-1948 Palestinian Press: Unsupervised Keyphrase Extraction and Temporal Discourse Analysis from Five Historical Arabic Newspapers. Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3oaq8zxp9hwx

@inproceedings{barakat2026mining,
  title = {{Mining the Pre-1948 Palestinian Press: Unsupervised Keyphrase Extraction and Temporal Discourse Analysis from Five Historical Arabic Newspapers}},
  author = {Basel Barakat and Nizam Barakat},
  booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026},
  year = {2026},
  doi = {10.63317/3oaq8zxp9hwx},
  url = {https://doi.org/10.63317/3oaq8zxp9hwx}
}

References

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

  1. Mohammed A. Al Ghamdi. 2022. A novel approach to printed Arabic optical character recognition. Arabian Journal for Science and Engineering, 47:2219–2237. find
  2. Jawad H Al-Khateeb, Jinchang Ren, Jianmin Jiang, and Husni Al-Muhtaseb. 2011. Offline handwritten Arabic cursive text recognition: A survey. Pattern Recognition, 44(3):555–567. find
  3. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. AraBERT: Transformer-based model for Arabic language understanding. In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, pages 9–15. ↗
  4. Ami Ayalon. 1995. The Press in the Arab Middle East: A History. Oxford University Press. ↗
  5. Ricardo Campos, Vítor Mangaravite, Arian Pasquali, Alípio Jorge, Célia Nunes, and Adam Jatowt. 2020. YAKE! keyword extraction from single documents using multiple local features. Information Sciences, 509:257–289. ↗
  6. Jaime Carbonell and Jade Goldstein. 1998. The use of MMR, diversity-based reranking for reordering documents and producing summaries. In Proceedings of the 21st Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 335– 336. find
  7. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8440–8451. find
  8. 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, pages 4171–4186.
  9. Amina El Ganadi, Luca Gagliardelli, Sania Aftar, and Federico Ruozzi. 2025. Digital Maktaba project: Proposing a metadata-driven framework for Arabic library digitization. In Proceedings of the 21st Conference on Information and Research Sciences Connecting to Digital and Library Science (IRCDL 2025), volume 3937 of CEUR Workshop Proceedings. find
  10. Safiullah Faizullah, Muhammad Shahid Ayub, Sajid Hussain, and Muhammad Arif Khan. 2025. Advancements and challenges in Arabic optical character recognition: A comprehensive survey. ACM Computing Surveys, 58(4):1–37. find
  11. Google Cloud. 2023. Cloud Vision API. https://cloud.google.com/vision. ↗
  12. David Graff and Mohamed Maamouri. 2006. Arabic Gigaword, third edition. Technical report, Linguistic Data Consortium. ↗
  13. Till Grallert. 2021. Catch me if you can! Approaching the Arabic press of the late Ottoman Eastern Mediterranean through digital history. Geschichte und Gesellschaft, 47(1):58–89. ↗
  14. Maarten Grootendorst. 2020. KeyBERT: Minimal keyword extraction with BERT. https://github.com/MaartenGr/KeyBERT. ↗
  15. William L Hamilton, Jure Leskovec, and Dan Jurafsky. 2016. Diachronic word embeddings reveal statistical laws of semantic change. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1489– 1501.
  16. Jrayed. Jrayed: Arabic newspaper archive of ottoman and mandatory palestine. Accessed: 2026-03-30. find
  17. Rashid Khalidi. 1997. Palestinian Identity: The Construction of Modern National Consciousness. Columbia University Press. find
  18. Thomas Lansdall-Welfare, Saatviga Sudhahar, James Thompson, Justin Lewis, and Nello Cristianini. 2017. Content analysis of 150 years of British periodicals. Proceedings of the National Academy of Sciences, 114(4):E457–E465. ↗
  19. Krystyna K. Matusiak and Qasem Abu Harb. 2011. Digitizing the historical periodical collection at the Al-Aqsa Mosque library in East Jerusalem. In Hartmut Walravens, editor, Newspapers: Legal Deposit and Research in the Digital Era, pages 271–290. De Gruyter Saur, Berlin. ↗
  20. Palestine Square. 1932. Caricature: Balfour and the woes his ill-fated promise brought to palestine. Accessed: 2026-03-30. find
  21. David A Smith, Ryan Cordell, and Abby Mullen. 2015. Computational methods for uncovering reprinted texts in antebellum newspapers. American Literary History, 27(3):E1–E15. ↗
  22. Nabil Wagaa and Hassene Kallel. 2023. Analysis of recent deep learning techniques for Arabic handwritten-text OCR and post-OCR correction. Applied Sciences, 13(13):7568. find
  23. Wikipedia contributors. 2025a. Al-karmil (newspaper) — Wikipedia, the free encyclopedia. https://en.wikipedia.org/wiki/Al-Karmil_(newspaper). [Online; accessed 27-February-2026]. ↗
  24. Wikipedia contributors. 2025b. Lisan al arab (newspaper) — Wikipedia, the free encyclopedia. https://en.wikipedia.org/wiki/Lisan_al-Arab_(newspaper). [Online; accessed 27-February-2026]. ↗
  25. Wikipedia contributors. 2026a. Al-difa’ — Wikipedia, the free encyclopedia. https://en.wikipedia.org/wiki/Al-Difa%27. [Online; accessed 27-February-2026]. ↗
  26. Wikipedia contributors. 2026b. Falastin — Wikipedia, the free encyclopedia. https://en.wikipedia.org/wiki/Falastin. [Online; accessed 27-February-2026]. ↗
  27. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaudhary, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, et al. 2020. Transformers: State-ofthe-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/barakat-2026-mining-pre-1948-palestinian-press/