--- title: "PalNLP at AR-MS Shared Task — paper by Ayesh (2026)" description: "Handwritten text recognition (HTR) for Arabic needs ground-truth transcriptions that reflect what is actually on the page, not a… 2 references." authors: ["Mutaz Ayesh"] year: 2026 venue: "Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026" source: "https://aclanthology.org/2026.nakbanlp-1.40/" doi: "10.63317/3e9dvjxc2gbx" licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)" licence_url: "https://creativecommons.org/licenses/by-nc/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/ayesh-2026-palnlp-ar-ms-shared-task/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ayesh-2026-palnlp-ar-ms-shared-task/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/39/) / PalNLP at AR-MS Shared Task: Guidelines Paper for Arabic Manuscript Transcription # PalNLP at AR-MS Shared Task: Guidelines Paper for Arabic Manuscript Transcription **Authors:** [Mutaz Ayesh](https://md.research.elamur.ai/free/en/authors/mutaz-ayesh/) · **Year:** 2026 · **Venue:** [Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026](https://md.research.elamur.ai/free/en/venues/nakbanlp/) ([International Workshop on Nakba Narratives as Language Resources @ LREC 2026](https://md.research.elamur.ai/free/en/venues/nakbanlp/2026/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.nakbanlp-1.40/) · [PDF](https://aclanthology.org/2026.nakbanlp-1.40.pdf) · [DOI 10.63317/3e9dvjxc2gbx](https://doi.org/10.63317/3e9dvjxc2gbx) · [OpenAlex](https://openalex.org/W7166516173) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary Handwritten text recognition (HTR) for Arabic needs ground-truth transcriptions that reflect what is actually on the page, not a cleaned-up version. This paper documents the transcription rules the PalNLP team, a single experienced annotator, used for Subtask 1 of the AR-MS (Arabic Manuscript Understanding) Shared Task at Nakba-NLP 2026. The batch was 500 line-level image crops from the Omar Al-Saleh Memoir Collection, dating from 1953 to 1965. The material is mostly handwritten, with a few typed lines. The organizers also supplied full-page context. The core principle is strict character-level fidelity. The annotator kept the following as written: - spelling and typos, - Arabic-Indic digits, - the undotted ى, hamza forms and diacritics, - unusual spacing, - truncated words at line edges. The paper also gives rules for hard cases: illegible segments, foreign names, and crops spanning several visual lines. Consistency was maintained through self-review and repeated agreement checks on CodaBench. The result is a realistic benchmark slice that keeps orthographic variation, so OCR/HTR systems are tested under real manuscript conditions. ## Key points - The batch has 500 lines, 38,252 characters, 6,757 words, 4,118 unique words, an average of 13.51 words per line, and 55 flagged edge cases. - Nothing was normalized: - Arabic-Indic digits were not converted to 0–9. - The author's typos and missing or extra hamza/madda were left in. - Word-final ى was kept where standard spelling requires ي. - Truncated words at crop edges were transcribed, using the tatwīl character when the visible letter shape required it. Spaces before punctuation and wide gaps between poem verses were also preserved. - For lines spanning two visual rows, a newline was inserted. Numbers written beneath the noun they modify (e.g., a time under "the hour") were placed inline right after that word. - Illegible segments were given their closest visual reading, then underlined and tagged "incomprehensible". Google Search was used only for unclear foreign words and named entities, such as French politicians and Palestinian figures. ## Interesting ideas - External lookups only proposed spelling hypotheses, and a reading was accepted only when the visual strokes fully confirmed it, so outside knowledge could never overwrite the source. - Every output was explicitly split into confident readings versus best-guess approximations, with an annotation, so later users know which parts to trust. - Strict fidelity to the original (keeping typos, nonstandard spellings and odd spacing) makes the data reflect real conditions instead of an idealized form. - A lone annotator stayed consistent by revisiting earlier decisions once recurring patterns became clear and by periodically checking agreement against an expert reference. - Wider page context was used to resolve ambiguous local units, but the unit itself still decided the final reading. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (world): Wikidata/WordNet candidates for an entity should be treated as hypotheses, kept only when the text's own evidence (mentions, attributes, context) supports them. - **I2** (learn): silver trees and frames could carry a per-span "approximate" flag, letting training down-weight or skip uncertain pieces instead of mixing them with confident data. - **I3** (tokens): the tokenizer and lemmatizer could store the raw surface form alongside the normalized lemma, so errors stay visible and explainable rather than silently corrected. - **I4** (learn): self-induced repair rules could trigger a re-pass over earlier outputs, with regular scoring against a held-out gold set to catch drift. - **I5** (frames): coreference and role assignment could consult neighbouring sentences for disambiguation, while refusing any reading the local UD tree contradicts. *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 - [Cardiff University](https://md.research.elamur.ai/free/en/universities/cardiff-university/) — University, Cardiff, United Kingdom · 22 works · 2,444 citations 📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works · [📚 Literature and Literary Theory](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/literature-and-literary-theory/) · 519 works 🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works ## Cite Cite as: Mutaz Ayesh. 2026. PalNLP at AR-MS Shared Task: Guidelines Paper for Arabic Manuscript Transcription. 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. ```bibtex @inproceedings{ayesh2026palnlp, title = {{PalNLP at AR-MS Shared Task: Guidelines Paper for Arabic Manuscript Transcription}}, author = {Mutaz Ayesh}, booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026}, year = {2026}, doi = {10.63317/3e9dvjxc2gbx}, url = {https://doi.org/10.63317/3e9dvjxc2gbx} } ``` - [← previous PAIR: A Pilot Dataset for Dual Perspective-based Video-Grounded Dialogue and Reconciliation](https://md.research.elamur.ai/free/en/papers/watson-2026-pair-pilot-dataset-dual-perspective/) - [next → Par-ITA: Benchmarking Seq2Seq and LLMs on a Human-Supervised Parallel Corpus for Italian Hyperpartisan Neutralization](https://md.research.elamur.ai/free/en/papers/maggini-2026-par-ita-benchmarking-seq2seq-llms/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/39/) ## References 2 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Hadi Hamoud, Ahmad Ali Chamseddine, Bilal Shalash, Firas Ben Abid, Mustafa Jarrar, Chadi Abou Chakra, Bernard Ghanem, and Fadi A. Zaraket. 2026. [find](https://scholar.google.com/scholar?q=Hadi+Hamoud%2C+Ahmad+Ali+Chamseddine%2C+Bilal+Shalash%2C+Firas+Ben+Abid%2C+Mustafa+Jarrar%2C+Chadi+Abou+Chakra%2C+Bernard+Ghanem%2C+and+Fadi+A%2E+Zaraket%2E+2026%2E) 2. [NAKBA NLP 2026: Shared Task on Arabic Handwritten Manuscript Understanding (Palestine Memory–Omar Al-Saleh Memoir)](https://md.research.elamur.ai/free/en/papers/hamoud-2026-nakba-nlp-2026-shared-task/). In Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources (Nakba-NLP 2026), co-located with the Language Resources and Evaluation Conference (LREC 2026), Palma, Mallorca, Spain. *Updated 1 November 2026*