Arabic NLP & Dialects
Datasets and models that understand Arabic as it's actually written online — across dialects, scripts, and code-switching.
- Dialect identification
- Low-resource NLP
- Arabizi
Research
Research & publications
I study how language technologies handle Arabic in all its variety, and how information — and misinformation — moves through Arabic-language media. Below are my core areas, the projects I'm leading now, and a complete list of publications.
Research areas
Misinformation & Media
How false claims spread across Arabic-language platforms, and why detection tools trained on English miss them.
- Fact-checking
- Network analysis
- Platform studies
AI Ethics & Governance
Auditing language technologies for bias, and shaping policy so the Arabic-speaking world isn't an afterthought.
- Algorithmic audits
- Content moderation
- Data governance
Digital Humanities & Archives
Computational methods for reading historical Arabic newspapers and preserving the region's born-digital record.
- OCR
- Web archiving
- Historical corpora
Current projects
Funded work in progress, with collaborators across four institutions.
LAHJA: An Open Corpus of Everyday Arabic
Building a 40-million-word, consent-based corpus of dialectal Arabic from 18 countries, with tools for researchers and startups.
- Funded by Regional Science Foundation
- 2024 – 2027
- With Northbridge University and the Maghreb Language Institute
Rumor Maps
Tracing how health misinformation moved through Arabic messaging apps during public-health crises, and testing interventions with fact-checkers.
- Funded by Open Information Fund
- 2025 – 2026
- With three regional fact-checking newsrooms
The Reading Room
OCR and search for 120,000 pages of Levantine newspapers (1908–1948), now freely available to historians.
- Funded by Heritage Futures Programme
- 2021 – 2024
- With the National Library digitization unit
Publications
Search by title, co-author, or venue. Copy a BibTeX entry with one click.
Showing 10 of 10
2026
Journal articleDialect-Aware Misinformation Detection for Arabic Social Media
L. Haddad, O. Nasser, J. Kim
Journal of Computational Media Studies, 12(2)
DOICited by 41Misinformation classifiers for Arabic are overwhelmingly trained on Modern Standard Arabic, yet most viral claims circulate in dialect. We introduce a benchmark of 38,000 fact-checked claims across nine dialect groups and show that dialect-aware pre-training reduces false negatives by 31% relative to strong multilingual baselines.
2025
Conference paperBest PaperLAHJA: A Consent-Based Corpus of Everyday Written Arabic
L. Haddad, S. Benali, R. Aziz, M. Farouk
Proceedings of the Arabic NLP Conference (ArabicNLP 2025)
DOICited by 87We describe the design and first release of LAHJA, a 12-million-word corpus of dialectal Arabic contributed by 4,100 volunteers with explicit, revocable consent. We detail our collection protocol, annotation scheme, and baseline results for dialect identification and sentiment analysis.
2025
Journal articleالتحيّز اللهجي في النماذج اللغوية الكبيرة: دراسة تجريبية على خمس لهجات عربية
ل. حدّاد، ع. الخطيب
مجلة اللسانيات العربية الحاسوبية، المجلد 7، العدد 1
DOICited by 23تقيس هذه الدراسة أداء خمسة نماذج لغوية كبيرة على مهام الفهم والتوليد بخمس لهجات عربية، وتُظهر فجوة في الأداء تصل إلى 24 نقطة بين الفصحى واللهجات المغاربية. نقترح إطارًا للتقييم العادل بين اللهجات ونناقش أثره على الخدمات الرقمية الحكومية.
2024
Conference paperAuditing Content Moderation in Arabic: Evidence from Three Platforms
L. Haddad, T. Ofori, D. Mansour
Proceedings of the ACM Conference on Fairness in Computing (FairComp '24)
DOICited by 112Using 60,000 paired posts, we measure how automated moderation systems treat equivalent content written in English, Modern Standard Arabic, and Levantine Arabic. Dialectal posts were removed at 2.3× the rate of their English equivalents, with disproportionate impact on news and advocacy accounts.
2024
Journal articleReading the Levantine Press at Scale: OCR, Search, and the Historian's Workflow
L. Haddad, N. Sayegh
Digital Scholarship in the Humanities, 39(4)
DOICited by 34We report on a three-year collaboration to digitise 120,000 pages of Levantine newspapers (1908–1948). Beyond character error rates, we examine how historians actually search and cite OCR'd sources, and propose design principles for archives serving Arabic-script collections.
2023
Book chapterLanguage Technology and the Politics of the Standard
L. Haddad
In A. Rahman & P. Cole (Eds.), The Handbook of Language and Digital Society. Meridian Academic Press
DOICited by 58This chapter traces how decisions about which variety of Arabic to support — in keyboards, spell-checkers, and machine translation — encode older debates about authenticity, nationhood, and modernity, and argues for a pluralist approach to language technology design.
2023
Journal articleانتشار الشائعات الصحية عبر تطبيقات المراسلة: تحليل شبكي لمجموعات عربية مفتوحة
ل. حدّاد، ي. العمري، هـ. سليمان
المجلة العربية لبحوث الإعلام والاتصال، العدد 42
DOICited by 45تحلل الدراسة 1.2 مليون رسالة في 340 مجموعة مراسلة عامة ناطقة بالعربية، وتحدد أنماط انتشار الادعاءات الصحية الزائفة ودور «العُقد الجسرية» في نقلها بين المجتمعات. تقدم النتائج توصيات عملية لمنصات التحقق.
2022
Conference paperBeyond Script: Normalising Arabizi for Downstream NLP
R. Aziz, L. Haddad
Findings of the Conference on Empirical Methods in Language Processing
DOICited by 96Arabizi — Arabic written in Latin characters and numerals — remains widespread yet poorly supported. We release a parallel Arabizi–Arabic corpus and a character-level transliteration model that improves downstream sentiment classification by 14 F1 points.
2021
BookWritten Voices: Variation and Identity in Online Arabic
L. Haddad
Northbridge University Press
Cited by 214Drawing on a decade of data from forums, blogs, and social media, this book shows how Arabic speakers use spelling, script, and dialect online to signal identity and belonging — and what this means for the technologies that attempt to read them.
2019
Journal articleFine-Grained Dialect Identification with Hierarchical Classifiers
L. Haddad, K. Lindqvist
Transactions on Language Resources, 7
DOICited by 305We present a hierarchical approach to identifying Arabic dialects at the city level, combining regional and local classifiers. The model reaches 71% accuracy across 25 cities and remains a widely used baseline.