Elamur Research⌕
☰
home / papers / 2026 / The NakbaEcho Dataset: From Oral Testimonies to a Transcribed Arabic History Corpus

The NakbaEcho Dataset: From Oral Testimonies to a Transcribed Arabic History Corpus

Paper · 2026 · Batool Najeh Balah, Mahmoud Fawzi, Houda Elmimouni et al.

Summary

Palestinian oral testimonies about the 1948 Nakba exist mostly as scattered audio and video recordings. Very little of this material is transcribed in a form that NLP or corpus research can use. Earlier computational work used small English-translated subsets or transcripts prepared for a single study. The authors built NakbaEcho to fill this gap.

They started from the Palestine Remembered Oral History index, which links to sources such as POHA and YouTube. They unified the interview metadata and ran an LLM-based (Gemini) transcription pipeline tuned for elderly speakers of Palestinian dialect. The output is timestamped, speaker-diarized Arabic text with extra annotations: speaker gender and emotion, plus chunk- and segment-level flags for mentions of the Nakba, Jews/Israel and the British Mandate.

The released corpus covers about 2,180 hours of recordings. The paper also evaluates transcription quality and describes the corpus by geography, narrator demographics and lexical patterns.

The resource matters for two reasons. It is a rare large source of naturally spoken dialectal Arabic, and it is a searchable archive of historical memory for digital humanities.

Key points

  • The index holds 1,341 interview entries, more than 3,000 hours in total, recorded from 1994 onward. Single recordings range from 10 minutes to over 25 hours. Of these, 708 interviews (about 2,180 h) were transcribed and released on GitHub (batouln/NakbaEcho), with more promised later.

  • The corpus covers 239 villages, and 94 of them appear in only one interview. Safed and Haifa together account for 50.1% of the interviews.

  • There are 719 narrators: 536 male (74.5%) and 183 female. Female share varies sharply by source, from 44.1% in POHA to 8.7% in YouTube interviews. The oldest narrator was born in 1897 and the youngest in 1962; the median birth year is 1929.

  • Gender was labelled in two independent ways:

    • from narrator names, using Claude Opus plus manual review of ambiguous names;
    • from voice, using Gemini during diarization.

    The two agree on 98.6% of the 579 comparable cases.

  • Words that differ by speaker gender were ranked with log-likelihood G² and a Bonferroni correction. Women's speech favours family and home vocabulary. Men's speech favours place, distance and measurement terms.

Interesting ideas

  • I1. Gender labels were validated by comparing two methods built on unrelated signals (name morphology vs. acoustics), so their agreement is real evidence rather than shared bias.
  • I2. The transcription prompt forbids guessing: unclear audio becomes [inaudible], and suspected place names are kept as heard and tagged [PLACE?] instead of being "corrected" into common words.
  • I3. Each diarized speaker gets a short literal voice description, which lets speaker IDs that are only consistent within one chunk be matched across chunks.
  • I4. Mention fields must be strictly literal summaries of what was said, and must be an empty string when the topic is absent, which separates detection from interpretation.
  • I5. Contrastive vocabulary between subgroups was found with G² plus Bonferroni correction rather than raw frequency differences.

For Mova

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

  • I1 (learn): When building silver data, prefer witness parsers or taggers with genuinely different feature sources, and count agreement only between independent witnesses.
  • I2 (tokens): The tagger and lemmatizer could emit an explicit "possible proper noun" flag for unknown capitalized or rare tokens instead of forcing them into a dictionary lemma.
  • I3 (frames): Coreference over long texts processed in windows could carry a compact descriptor per entity (role, attributes, typical predicates) to merge mentions across window boundaries.
  • I4 (world): World-building could store only explicitly stated facts in one layer and keep inferred ones in a separate, marked layer, so every state stays traceable to text.
  • I5 (learn): The same test could compare parser outputs or error profiles across genres or treebanks and flag only statistically solid differences as candidates for repair rules.

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

🎓University of EdinburghEdinburgh · United Kingdom359 works · 77,352 citations🎓University of ManitobaWinnipeg · Canada5 works · 2,378 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Batool Najeh Balah, Mahmoud Fawzi, Houda Elmimouni and Walid Magdy. 2026. The NakbaEcho Dataset: From Oral Testimonies to a Transcribed Arabic History Corpus. 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 10.63317/4zvtrpg8sm2s

@inproceedings{balah2026nakbaecho,
  title = {{The NakbaEcho Dataset: From Oral Testimonies to a Transcribed Arabic History Corpus}},
  author = {Batool Najeh Balah, Mahmoud Fawzi, Houda Elmimouni and Walid Magdy},
  booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026},
  year = {2026},
  doi = {10.63317/4zvtrpg8sm2s},
  url = {https://doi.org/10.63317/4zvtrpg8sm2s}
}

References

23 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. American University of Beirut. Palestinian oral history archive (poha). https://www.aub.edu.lb/ifi/Pages/poha.aspx. Accessed: 2026-02-27. ↗
  2. Huthaifa I Ashqar. 2025. Sentiment analysis of nakba oral histories: A critical study of large language models. In Proceedings of the first International Workshop on Nakba Narratives as Language Resources, pages 30–36. ↗
  3. Ghadir A Awad, Tamara N Rayan, Lavinia Dunagan, and David Gamba. 2025. Collective memory and narrative cohesion: A computational study of palestinian refugee oral histories in lebanon. In Proceedings of the first International Workshop on Nakba Narratives as Language Resources, pages 83–102. find
  4. Amirbek Djanibekov, Hawau Olamide Toyin, Raghad Alshalan, Abdullah Alatir, and Hanan Aldarmaki. 2025. Dialectal coverage and generalization in arabic speech recognition. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 29490–29502. ↗
  5. Encyclopaedia Britannica. 2026. 1948 arab– israeli war. https://www.britannica.com/event/1948-Arab-Israeli-War. Accessed: 2026-02-27. ↗
  6. Osama Hamed and Nadeem Zaidkilani. 2025. Arabic topic classification corpus of the nakba short stories. In Proceedings of the first International Workshop on Nakba Narratives as Language Resources, pages 48–55.
  7. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. 2021. The interplay of variant, size, and task type in arabic pre-trained language models.
  8. Mustafa Jarrar, Nizar Habash, Mo El-Haj, Amal Haddad Haddad, Zeina Jallad, Camille Mansour, Diana Allan, Paul Rayson, Tymaa Hammouda, and Sanad Malaysha. 2025. Proceedings of the first international workshop on nakba narratives as language resources. In Proceedings of the first International Workshop on Nakba Narratives as Language Resources. find
  9. Fatma Kassem. 2011. Palestinian women: Narrative histories and gendered memory. Bloomsbury Publishing. find
  10. Amr Keleg, Sharon Goldwater, and Walid Magdy. 2023. Aldi: Quantifying the arabic level of dialectness of text. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10597–10611.
  11. Walid Khalidi. 1992. All that remains. The Palestinian villages occupied and. find
  12. Annie K Lamar, Rick Castle, Carissa Chappell, Emmanouela Schoinoplokaki, Allene M Seet, Amit Shilo, and Chloe Nahas. 2025. Cognitive geographies of catastrophe narratives: Georeferenced interview transcriptions as language resource for models of forced displacement. In Proceedings of the first International Workshop on Nakba Narratives as Language Resources, pages 18–29. find
  13. Nur Masalha. 2008. Remembering the palestinian nakba: Commemoration, oral history and narratives of memory. Holy Land Studies, 7(2):123– 156. ↗
  14. Masalha Nur. 2008. Remembering the palestinian nakba: Commemoration, oral history and narratives of memory. Holy Land Studies, 7(2):123– 156. ↗
  15. Ahmad H Sa’di and Lila Abu-Lughod. 2007. Nakba: Palestine, 1948, and the claims of memory. Columbia University Press. find
  16. Rosemary Sayigh. 1994. Too Many Enemies: The Palestinian Experience in Lebanon. Zed Books, London. find
  17. Hana Sleiman and Kaoukab Chebaro. 2018. Narrating palestine: The palestinian oral history archive project. Journal of Palestine Studies, 47(2):63– 76. ↗
  18. Bashar Talafha, Karima Kadaoui, Samar Mohamed Magdy, Mariem Habiboullah, Chafei Mohamed Chafei, Ahmed Oumar El-Shangiti, Hiba Zayed, Mohamedou Cheikh Tourad, Rahaf Alhamouri, Rwaa Assi, et al. 2024. Casablanca: Data and models for multidialectal arabic speech recognition. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21745–21758. ↗
  19. The Nakba Archive. 2002. Nakba archive. https://www.nakba-archive.org/. Accessed: 2026-02-27. ↗
  20. United Nations Committee on the Exercise of the Inalienable Rights of the Palestinian People. 2025. About the nakba. https://www.un.org/unispal/about-the-nakba/. Accessed: 2026-02-27. ↗
  21. United Nations General Assembly. 1947. Future government of palestine: United nations general assembly resolution 181 (ii). https://digitallibrary.un.org/Adopted 29 November record/210008. 1947. ↗
  22. United Nations General Assembly. 1948. Palestine—progress report of the united nations mediator: United nations general assembly resolution 194 (iii). https://www.unrwa.org/content/resolution-194. Adopted 11 December 1948. ↗
  23. United Nations Relief and Works Agency for Palestine Refugees in the Near East. 2026. Palestine refugees. https://www.unrwa.org/palestine-refugees. Accessed: 2026-02-27. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/balah-2026-nakbaecho-dataset-oral-testimonies-transcribed/