---
title: "Helpful or Harmful? The Dual Role of Linguistic Features in"
description: "The paper asks whether explicit linguistic annotation helps large language models translate between Algerian… 56.08 COMET. 44 references."
authors: ["Abdelhalim Hafedh Dahou", "Mohamed Amine Cheragui"]
year: 2026
venue: "The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks"
source: "https://aclanthology.org/2026.osact-1.8/"
doi: "10.63317/5dpmbr8bbedw"
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/dahou-2026-helpful-or-harmful-dual-role/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/dahou-2026-helpful-or-harmful-dual-role/)

[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/32/) / Helpful or Harmful? The Dual Role of Linguistic Features in LLM-Based Dialectal Machine Translation

# Helpful or Harmful? The Dual Role of Linguistic Features in LLM-Based Dialectal Machine Translation

**Authors:** [Abdelhalim Hafedh Dahou](https://md.research.elamur.ai/free/en/authors/abdelhalim-hafedh-dahou/) and [Mohamed Amine Cheragui](https://md.research.elamur.ai/free/en/authors/mohamed-amine-cheragui/) · **Year:** 2026 · **Venue:** [The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks](https://md.research.elamur.ai/free/en/venues/osact/) ([Workshop on Open-Source Arabic Corpora and Processing Tools with Shared Tasks 2026](https://md.research.elamur.ai/free/en/venues/osact/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **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.osact-1.8/) · [PDF](https://aclanthology.org/2026.osact-1.8.pdf) · [DOI 10.63317/5dpmbr8bbedw](https://doi.org/10.63317/5dpmbr8bbedw) · [OpenAlex](https://openalex.org/W7166593726)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks whether explicit linguistic annotation helps large language models translate between Algerian Arabic (Darija) and Modern Standard Arabic (MSA). The authors took a linguistically enriched subset of the PADIC corpus and ran translation in both directions under three prompt conditions: plain text, text with POS tags, and text with diacritics. They tested a broad set of frontier systems (GPT-4o, GPT-4-1106, several Claude versions, Gemini 1.5 Pro/Flash) and two open-weight models (Llama3.1-8B, Gemma2-9B). Evaluation combined COMET, BERT-based similarity, BLEU and ChrF++ with human ratings of adequacy and fluency.

The effects were mixed and depended on direction. Diacritics sometimes raised human-judged adequacy when translating into the dialect. POS tags and forced diacritization mostly added noise, and the damage was worst when translating from dialect into MSA. Automatic overlap metrics often disagreed with human judgments. The authors conclude that adding linguistic features is not a reliable improvement and should be tested case by case. This is a useful caution for any pipeline that feeds structured annotations to an LLM.

## Key points
- **Setup:** bidirectional ALG↔MSA translation on a PADIC subset, under three conditions: no features, POS tags, and diacritics. Code, prompts and data are released in the FeatureDual-MT repository.
- **ALG→MSA with diacritics:** every model and every metric dropped. For example, Claude-3.5-sonnet-10 fell from 79.19 to 56.08 COMET and from 9.02 to 3.39 BLEU.
- **MSA→ALG, automatic vs. human scores:** with diacritics, COMET fell for almost all models (Gemma2-9B was the exception). Yet human adequacy rose for GPT-4 (88.63→92.95) and Gemini 1.5 Pro (91.46→94.53).
- **POS tags in human evaluation:** POS tags lowered fluency in every case. Adequacy collapsed for GPT-4 in MSA→ALG (88.63→34.74), and the only adequacy gain was Gemini 1.5 Pro in ALG→MSA (52.32→61.68).
- **Model gap and error types:**
  - Small open-weight models scored near zero in human evaluation.
  - Frontier models understood the dialect well but generated it less reliably.
  - Generation errors included:
    - keeping MSA vocabulary instead of dialect words;
    - slipping into Moroccan words;
    - character-level misspellings;
    - gender-agreement mistakes.

## Interesting ideas
- Measure the value of an added annotation separately for each direction or task, because the same cue helped one direction and hurt the other.
- Dense or unfamiliar annotation (forced full diacritization) acts as distribution-shifted input, and models then produce fluent text that ignores the source.
- POS-augmented prompts caused a distinctive error: subject and object roles were swapped while the output still read fluently.
- Overlap metrics and even COMET moved opposite to human adequacy, so string similarity alone is a weak judge of meaning preservation.
- The error taxonomy flags "variety leakage": output drifts into a neighbouring dialect or stays in the source variety.

## For Mova

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

- **I1** (learn): When LLM teacher tiers receive UD tags or other annotations in their prompts, run a with/without ablation per task rather than assuming the extra structure always helps.
- **I2** (tokens): Expose only the morphological features that actually resolve an ambiguity, rather than annotating every token, when handing tagged text to downstream components or teachers.
- **I3** (frames): Comparing who-did-what-to-whom frames of source and output would catch exactly these role inversions, which fluency and overlap scores miss.
- **I4** (back): Score the text → graph → text compressor by comparing the meaning graphs of input and regenerated text, and keep a small human check, rather than relying on BLEU/ChrF-style overlap.
- **I5** (back): Add a target-register lexicon check after generation, so the output stays in the intended variety or style and is not merely copied from the input's.

*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

- [GESIS - Leibniz Institute for the Social Sciences](https://md.research.elamur.ai/free/en/labs/gesis-leibniz-institute-for-the-social-sciences/) — Research institute, Mannheim, Germany · 10 works · 5 citations
- [Ahmed Draia University](https://md.research.elamur.ai/free/en/universities/ahmed-draia-university/) — University, Adrar, Algeria · 2 works · 1 citation

📚 Shelves (4): [🖥️ 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

🧭 Topics (3): [🔁 Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/) · 1,087 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🛡️ Bias, fairness & safety](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/) · 517 works

## Cite

Cite as: Abdelhalim Hafedh Dahou and Mohamed Amine Cheragui. 2026. Helpful or Harmful? The Dual Role of Linguistic Features in LLM-Based Dialectal Machine Translation. The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{dahou2026helpful,
  title = {{Helpful or Harmful? The Dual Role of Linguistic Features in LLM-Based Dialectal Machine Translation}},
  author = {Abdelhalim Hafedh Dahou and Mohamed Amine Cheragui},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks},
  year = {2026},
  doi = {10.63317/5dpmbr8bbedw},
  url = {https://doi.org/10.63317/5dpmbr8bbedw}
}
```

- [← previous HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model](https://md.research.elamur.ai/free/en/papers/kayzer-2026-hebatron-hebrew-specialized-open-weight/)
- [next → Hemolix.TabGen: Optimized Table Generation from Documents](https://md.research.elamur.ai/free/en/papers/shrestha-2026-hemolix-tabgen-optimized-table-generation/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

44 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ahmed Elmogtaba Abdelmoniem Ali Abdelaziz, Ashraf Hatim Elneima, and Kareem Darwish. 2024. Llm-based mt data creation: Dialectal to msa translation shared task. In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation@ LREC-COLING 2024, pages 112–116. [find](https://scholar.google.com/scholar?q=Ahmed+Elmogtaba+Abdelmoniem+Ali+Abdelaziz%2C+Ashraf+Hatim+Elneima%2C+and+Kareem+Darwish%2E+2024%2E+Llm%2Dbased+mt+data+creation%3A+Dialectal+to+msa+translation+shared+task%2E+In+Proceedings+of+the+6th+Workshop+on+O)
2. Muhammad Abdul-Mageed and Houda Bouamor. 2021. Arabic text processing with morphological and diacritical features. Computational Linguistics. [find](https://scholar.google.com/scholar?q=Muhammad+Abdul%2DMageed+and+Houda+Bouamor%2E+2021%2E+Arabic+text+processing+with+morphological+and+diacritical+features%2E+Computational+Linguistics%2E)
3. Muhammad Abdul-Mageed, AbdelRahim Elmadany, et al. 2021. [Arbert & marbert: Deep bidirectional transformers for arabic](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/). In Proceedings of the 59th annual meeting of the association for computational linguistics and the 11th international joint conference on natural language processing (volume 1: long papers), pages 7088–7105. [↗](https://doi.org/10.48550/arxiv.2101.01785)
4. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. [Gpt-4 technical report](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/). arXiv preprint arXiv:2303.08774. [↗](https://arxiv.org/abs/2303.08774)
5. Abdullah Alabdullah, Lifeng Han, and Chenghua Lin. 2025. [Advancing dialectal arabic to modern standard arabic machine translation](https://md.research.elamur.ai/free/en/papers/alabdullah-2025-advancing-dialectal-arabic-modern-standard/). arXiv preprint arXiv:2507.20301. [↗](https://arxiv.org/abs/2507.20301)
6. Sawsan Alqahtani, Mahmoud Ghoneim, and Mona Diab. 2016. Investigating the impact of various partial diacritization schemes on arabic-english statistical machine translation. In Conferences of the Association for Machine Translation in the Americas: MT Researchers’ Track, pages 191– 204. [find](https://scholar.google.com/scholar?q=Sawsan+Alqahtani%2C+Mahmoud+Ghoneim%2C+and+Mona+Diab%2E+2016%2E+Investigating+the+impact+of+various+partial+diacritization+schemes+on+arabic%2Denglish+statistical+machine+translation%2E+In+Conferences+of+the+Asso)
7. AI Anthropic. 2024. The claude 3 model family: Opus, sonnet, haiku. Claude-3 Model Card, 1. [find](https://scholar.google.com/scholar?q=AI+Anthropic%2E+2024%2E+The+claude+3+model+family%3A+Opus%2C+sonnet%2C+haiku%2E+Claude%2D3+Model+Card%2C+1%2E)
8. Hanin Atwany, Nour Rabih, Ibrahim Mohammed, Abdul Waheed, and Bhiksha Raj. 2024. [Osact 2024 task 2: Arabic dialect to msa translation](https://md.research.elamur.ai/free/en/papers/atwany-2024-osact-2024-task-2-arabic/). In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation@ LREC-COLING 2024, pages 98–103.
9. Baligh Babaali, Mohammed Salem, and Nawaf R Alharbe. 2025. Breaking language barriers with chatgpt: enhancing low-resource machine translation between algerian arabic and msa. International Journal of Information Technology, 17(7):4109–4118. [↗](https://doi.org/10.1007/s41870-024-01926-7)
10. Laith H Baniata, Seyoung Park, and Seong-Bae Park. 2018. A multitask-based neural machine translation model with part-of-speech tags integration for arabic dialects. Applied Sciences, 8(12):2502. [↗](https://doi.org/10.3390/app8122502)
11. Reem Bassiouney. 2009. Arabic Sociolinguistics. Edinburgh University Press. [find](https://scholar.google.com/scholar?q=Reem+Bassiouney%2E+2009%2E+Arabic+Sociolinguistics%2E+Edinburgh+University+Press%2E)
12. Mohamed Benrabah. 2014. Competition between four ”world” languages in algeria. Journal of World Languages, 1(1):38–59. [↗](https://doi.org/10.1080/21698252.2014.893676)
13. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, et al. 2018. The madar arabic dialect corpus and lexicon. In Proceedings of the eleventh international conference on language resources and evaluation (LREC 2018). [find](https://scholar.google.com/scholar?q=Houda+Bouamor%2C+Nizar+Habash%2C+Mohammad+Salameh%2C+Wajdi+Zaghouani%2C+Owen+Rambow%2C+Dana+Abdulrahim%2C+Ossama+Obeid%2C+Salam+Khalifa%2C+Fadhl+Eryani%2C+Alexander+Erdmann%2C+et+al%2E+2018%2E+The+madar+arabic+dialect+corpus)
14. Farida Bouhadiba. 2017. Algerian arabic: A sociolinguistic overview. In Elabbas Benmamoun and Reem Bassiouney, editors, The Routledge Handbook of Arabic Linguistics, pages 353–368. Routledge. [find](https://scholar.google.com/scholar?q=Farida+Bouhadiba%2E+2017%2E+Algerian+arabic%3A+A+sociolinguistic+overview%2E+In+Elabbas+Benmamoun+and+Reem+Bassiouney%2C+editors%2C+The+Routledge+Handbook+of+Arabic+Linguistics%2C+pages+353%E2%80%93368%2E+Routledge%2E)
15. Mohamed Amine Cheragui, Abdelhalim Hafedh Dahou, and Amin Abdedaiem. 2023. Exploring bert models for part-of-speech tagging in the algerian dialect: A comprehensive study. In Proceedings of the 6th International Conference on Natural Language and Speech Processing (ICNLSP 2023), pages 140–150. [find](https://scholar.google.com/scholar?q=Mohamed+Amine+Cheragui%2C+Abdelhalim+Hafedh+Dahou%2C+and+Amin+Abdedaiem%2E+2023%2E+Exploring+bert+models+for+part%2Dof%2Dspeech+tagging+in+the+algerian+dialect%3A+A+comprehensive+study%2E+In+Proceedings+of+the+6th+In)
16. Marta R Costa-Jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, et al. 2022. No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672. [↗](https://arxiv.org/abs/2207.04672)
17. Anna Currey and Kenneth Heafield. 2020. Incorporating linguistic annotations into neural machine translation. In Proceedings of WMT. [find](https://scholar.google.com/scholar?q=Anna+Currey+and+Kenneth+Heafield%2E+2020%2E+Incorporating+linguistic+annotations+into+neural+machine+translation%2E+In+Proceedings+of+WMT%2E)
18. Kareem Darwish and Hamdy Mubarak. 2016. [Farasa: A new fast and accurate arabic word segmenter](https://md.research.elamur.ai/free/en/papers/darwish-2016-farasa-new-fast-accurate-arabic/). In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 1070–1074.
19. Mona Diab, Mahmoud Ghoneim, and Nizar Habash. 2007. [Arabic diacritization in the context of statistical machine translation](https://md.research.elamur.ai/free/en/papers/diab-2007-arabic-diacritization-context-statistical-machine/). In Proceedings of Machine Translation Summit XI: Papers.
20. Mona Diab and Kadri Hacioglu. 2004. Automatic diacritization of arabic text. In Proceedings of COLING. [find](https://scholar.google.com/scholar?q=Mona+Diab+and+Kadri+Hacioglu%2E+2004%2E+Automatic+diacritization+of+arabic+text%2E+In+Proceedings+of+COLING%2E)
21. Lukas Edman and Anders Sogaard. 2023. Do large language models benefit from linguistic supervision? In Proceedings of ACL. [find](https://scholar.google.com/scholar?q=Lukas+Edman+and+Anders+Sogaard%2E+2023%2E+Do+large+language+models+benefit+from+linguistic+supervision%3F+In+Proceedings+of+ACL%2E)
22. Ashraf Hatim Elneima, AhmedElmogtaba Abdelmoniem Ali Abdelaziz, and Kareem Darwish. 2024. Osact6 dialect to msa translation shared task overview. In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation@ LREC-COLING 2024, pages 93–97. [find](https://scholar.google.com/scholar?q=Ashraf+Hatim+Elneima%2C+AhmedElmogtaba+Abdelmoniem+Ali+Abdelaziz%2C+and+Kareem+Darwish%2E+2024%2E+Osact6+dialect+to+msa+translation+shared+task+overview%2E+In+Proceedings+of+the+6th+Workshop+on+Open%2DSource+Arab)
23. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). arXiv preprint arXiv:2407.21783. [↗](https://arxiv.org/abs/2407.21783)
24. Nizar Habash. 2010. Introduction to Arabic Natural Language Processing. Morgan & Claypool. [↗](https://doi.org/10.1007/978-3-031-02139-8)
25. Nizar Habash. 2013. Arabic Natural Language Processing. Synthesis Lectures on Human Language Technologies. [find](https://scholar.google.com/scholar?q=Nizar+Habash%2E+2013%2E+Arabic+Natural+Language+Processing%2E+Synthesis+Lectures+on+Human+Language+Technologies%2E)
26. Nizar Habash. 2019. Arabic Natural Language Processing. Cambridge University Press. [find](https://scholar.google.com/scholar?q=Nizar+Habash%2E+2019%2E+Arabic+Natural+Language+Processing%2E+Cambridge+University+Press%2E)
27. Nizar Habash and Owen Rambow. 2007. [Arabic diacritization through full morphological tagging](https://md.research.elamur.ai/free/en/papers/habash-2007-arabic-diacritization-through-full-morphological/). In Proceedings of NAACL.
28. Clive Holes. 2004. Modern Arabic: Structures, Functions, and Varieties, revised edition. Georgetown University Press. [find](https://scholar.google.com/scholar?q=Clive+Holes%2E+2004%2E+Modern+Arabic%3A+Structures%2C+Functions%2C+and+Varieties%2C+revised+edition%2E+Georgetown+University+Press%2E)
29. 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](https://md.research.elamur.ai/free/en/papers/inoue-2021-interplay-variant-size-task-type/). arXiv preprint arXiv:2103.06678. [↗](https://arxiv.org/abs/2103.06678)
30. Salam Khalifa et al. 2022. Camel tools: An open platform for [arabic natural language processing](https://md.research.elamur.ai/free/en/papers/habash-2022-arabic-natural-language-processing/). In Proceedings of LREC.
31. Abdullah Salem Khered, Youcef Benkhedda, and Riza Theresa Batista-Navarro. 2025. [Dial2msa-verified: A multi-dialect arabic social media dataset for neural machine translation to modern standard arabic](https://md.research.elamur.ai/free/en/papers/khered-2025-dial2msa-verified-multi-dialect-arabic/). In Proceedings of the 4th Workshop on Arabic Corpus Linguistics (WACL-4), pages 50–62.
32. Djamel Eddine Kouloughli. 1999. Le maghrébin. Presses Universitaires de France. [find](https://scholar.google.com/scholar?q=Djamel+Eddine+Kouloughli%2E+1999%2E+Le+maghr%C3%A9bin%2E+Presses+Universitaires+de+France%2E)
33. Yinhan Liu and Graham Neubig. 2024. Revisiting explicit syntax in neural machine translation. Transactions of the ACL. [find](https://scholar.google.com/scholar?q=Yinhan+Liu+and+Graham+Neubig%2E+2024%2E+Revisiting+explicit+syntax+in+neural+machine+translation%2E+Transactions+of+the+ACL%2E)
34. Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas, and Kamel Smaili. 2015. [Machine translation experiments on padic: A parallel arabic dialect corpus](https://md.research.elamur.ai/free/en/papers/meftouh-2015-machine-translation-experiments-padic-parallel/). In Proceedings of the 29th Pacific Asia conference on language, information and computation, pages 26–34.
35. Rasha Obeidat, Luay Alawneh, and Yara Al-Harahsheh. 2025. Exploring prompting for dialectical machine translation: a focus on north jordanian arabic. PeerJ Computer Science, 11:e3209. [↗](https://doi.org/10.7717/peerj-cs.3209)
36. Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. [A universal part-of-speech tagset](https://md.research.elamur.ai/free/en/papers/petrov-2012-universal-part-speech-tagset/). In Proceedings of LREC.
37. Lotfi Sayahi. 2014. Diglossia and Language Contact: Language Variation and Change in North Africa. Cambridge University Press. [find](https://scholar.google.com/scholar?q=Lotfi+Sayahi%2E+2014%2E+Diglossia+and+Language+Contact%3A+Language+Variation+and+Change+in+North+Africa%2E+Cambridge+University+Press%2E)
38. Rico Sennrich and Barry Haddow. 2016. [Linguistic input features improve neural machine translation](https://md.research.elamur.ai/free/en/papers/sennrich-2016-linguistic-input-features-improve-neural/). In Proceedings of the First Conference on Machine Translation: Volume 1, Research Papers, pages 83–91.
39. Serry Sibaee, Omer Nacar, Yasser Al-Habashi, Adel Ammar, and Wadii Boulila. 2025. [Shami-mt: a syrian arabic dialect to modern standard arabic bidirectional machine translation system](https://md.research.elamur.ai/free/en/papers/sibaee-2025-shami-mt-syrian-arabic-dialect/). arXiv preprint arXiv:2508.02268. [↗](https://arxiv.org/abs/2508.02268)
40. Gemini Team, Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, et al. 2023. [Gemini: a family of highly capable multimodal models](https://md.research.elamur.ai/free/en/papers/gemini-team-2023-gemini-family-highly-capable-multimodal/). arXiv preprint arXiv:2312.11805. [↗](https://arxiv.org/abs/2312.11805)
41. Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, et al. 2024. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv:2408.00118. [↗](https://arxiv.org/abs/2408.00118)
42. Kees Versteegh. 2014. The Arabic Language, 2 edition. Edinburgh University Press. [find](https://scholar.google.com/scholar?q=Kees+Versteegh%2E+2014%2E+The+Arabic+Language%2C+2+edition%2E+Edinburgh+University+Press%2E)
43. Nasser Zalmout and Nizar Habash. 2020. Adversarial neural arabic diacritization. In Proceedings of ACL. [find](https://scholar.google.com/scholar?q=Nasser+Zalmout+and+Nizar+Habash%2E+2020%2E+Adversarial+neural+arabic+diacritization%2E+In+Proceedings+of+ACL%2E)
44. Imed Zitouni, Jeffrey Sorensen, and Ruhi Sarikaya. 2006. Arabic diacritization using a statistical approach. In Proceedings of HLT-NAACL. [find](https://scholar.google.com/scholar?q=Imed+Zitouni%2C+Jeffrey+Sorensen%2C+and+Ruhi+Sarikaya%2E+2006%2E+Arabic+diacritization+using+a+statistical+approach%2E+In+Proceedings+of+HLT%2DNAACL%2E)

*Updated 1 November 2026*