Building Arabic NLP from the Ground Up: Twenty Years of Lessons, Failures, and Open Problems

Wajdi Zaghouani

arxiv: 2605.20786 · v1 · pith:JDRB2MUMnew · submitted 2026-05-20 · 💻 cs.CL

Building Arabic NLP from the Ground Up: Twenty Years of Lessons, Failures, and Open Problems

Wajdi Zaghouani This is my paper

Pith reviewed 2026-05-21 05:14 UTC · model grok-4.3

classification 💻 cs.CL

keywords Arabic NLPlow-resource languagesdataset creationshared taskscomputational social scienceresearch infrastructureNLP failuresunderserved languages

0 comments

The pith

Twenty years of Arabic NLP work reveals that social, institutional, and epistemic barriers matter more than linguistic ones for underserved languages.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper reflects on two decades spent building foundational linguistic resources for Arabic and then shifting toward social media analysis and applied systems. It draws three lessons from the process: dataset creation is fundamentally a social coordination task, communities formed around shared tasks often outlast the tasks themselves, and moving into computational social science surfaces problems that standard NLP training does not prepare researchers to handle. By examining concrete failures, including an unused depression detection corpus and an overly broad spread across shared tasks, the author concludes that the primary obstacles for languages like Arabic are not technical but social and institutional.

Core claim

The central claim is that the hardest problems in developing NLP for underserved communities are not linguistic but social, institutional, and epistemic, and require competencies the field rarely teaches. This emerges from the contrast between the first decade of infrastructure building and the second decade of socially oriented applications, along with the specific failures that arose when social and institutional factors were not addressed.

What carries the argument

Reflective examination of three lessons and three failures drawn from the author's sequence of Arabic NLP projects, treated as evidence that non-technical factors dominate progress.

If this is right

Dataset building for low-resource languages requires explicit attention to social processes of collection and annotation.
Shared tasks produce durable research communities whose value exceeds the immediate technical outputs.
Traditional NLP training leaves gaps when researchers attempt to apply resources to real-world social problems.
Assumptions that infrastructure for a standard variety will transfer to dialects without targeted adaptation are likely to fail.
Deployment of NLP tools into practice settings is blocked more by institutional barriers than by model accuracy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Other low-resource language communities could accelerate progress by treating community formation as a primary goal rather than a side effect of shared tasks.
NLP training programs might add modules on institutional navigation and interdisciplinary collaboration to reduce similar failures elsewhere.
Prioritizing long-term social infrastructure over short-term task performance could produce more sustainable resources across many languages.
The pattern suggests value in pairing NLP groups with sociologists or ethicists from the start of resource-building projects.

Load-bearing premise

The author's experiences with specific Arabic projects are representative of the general challenges faced by other underserved languages and that the identified failures stem primarily from social and institutional factors rather than technical limitations.

What would settle it

A clinical deployment of an Arabic depression detection system that succeeds without changes to institutional or social processes, or a project showing clean transfer of Modern Standard Arabic infrastructure to dialectal tasks without additional social coordination.

read the original abstract

This paper reflects on twenty years of building NLP resources and research infrastructure for Arabic, a language spoken by hundreds of millions yet historically underserved relative to languages such as English or Chinese. The first decade focused on foundational linguistic infrastructure; the second shifted toward computational social science, social media analysis, and socially oriented applications. Rather than cataloguing outputs, the paper examines what the experience of building them revealed. Three counterintuitive lessons emerge: building datasets is as much a social process as a technical one; communities formed around shared tasks often matter more than the tasks themselves; and moving from language resources to computational social science exposes challenges that traditional NLP training does not address. We discuss three failures: a depression detection corpus that never reached clinical practice, a period of spreading across too many shared tasks without sufficient depth, and a long-standing assumption that Modern Standard Arabic infrastructure would transfer cleanly to dialectal tasks. These experiences suggest that the hardest problems in developing NLP for underserved communities are not linguistic but social, institutional, and epistemic, and require competencies the field rarely teaches.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

After twenty years the hardest parts of Arabic NLP turned out to be the social and institutional ones rather than the tech.

read the letter

The main takeaway is that social, institutional, and epistemic issues have been bigger barriers than technical or linguistic ones in developing Arabic NLP over the past twenty years. The paper lays out three lessons from the author's work: datasets involve social processes, task communities can outweigh the tasks, and shifting to computational social science brings new challenges not covered in standard training. It pairs these with three failures, including a depression detection corpus that stayed academic, spreading efforts too thin across shared tasks, and expecting Modern Standard Arabic tools to handle dialects without adjustment. These examples come from actual projects in the first decade on infrastructure and the second on applications. This approach works because it focuses on what the experience revealed rather than listing achievements. Being open about the failures gives readers something concrete to learn from, which is uncommon in the literature. The narrative structure makes it easy to follow the shift in focus over time. The limitation is the reliance on personal case studies without supporting data or comparisons. The attribution of problems to social factors is reasonable based on the stories told, but it lacks the kind of evidence that would make the generalization stronger. Other researchers might have encountered different primary obstacles depending on their resources and networks. This is aimed at NLP researchers working on low-resource or underserved languages who are thinking about long-term sustainability and community aspects. Someone looking for new methods or empirical results on Arabic processing won't get that here, but those interested in the human side of resource building could find it relevant. I would send this to peer review. It raises points worth discussing in the field even if it is more essay than research report.

Referee Report

2 major / 2 minor

Summary. The paper reflects on the author's twenty years of experience developing NLP resources and infrastructure for Arabic. It contrasts the first decade's focus on foundational linguistic tools with the second decade's emphasis on computational social science and socially oriented applications. Rather than listing outputs, it extracts three counterintuitive lessons (dataset creation as a social process, the value of communities over tasks, and the limits of traditional NLP training for social science work) and analyzes three failures (a depression-detection corpus that did not reach clinical use, over-spreading across shared tasks, and the assumption that Modern Standard Arabic resources would transfer to dialects). The central claim is that the hardest problems for underserved languages are social, institutional, and epistemic rather than linguistic.

Significance. If the interpretive lessons hold, the paper supplies a valuable practitioner perspective on why technical progress alone has not closed the gap for low-resource languages. It highlights competencies (community building, institutional navigation, epistemic humility) that standard NLP curricula rarely address and could usefully inform funding priorities, training programs, and evaluation criteria for work on other underserved languages.

major comments (2)

[three failures and concluding paragraph] The central claim that failures stem primarily from social and institutional factors rather than technical limitations is load-bearing for the paper's conclusion, yet it rests on narrative case studies without quantitative metrics, controlled comparisons to successful projects, or external validation. This weakens the causal attribution (see the three-failures discussion and the final paragraph).
[final paragraph and lessons section] The assertion that the author's experiences with Arabic are representative of challenges for other underserved languages is stated without supporting evidence or discussion of selection effects. A concrete test would be to compare the identified social barriers against documented cases for at least one additional language family.

minor comments (2)

[abstract] The abstract and introduction could more explicitly state the paper's genre (reflective essay rather than empirical study) to set reader expectations.
[failures discussion] Some project names and shared-task references are mentioned without citations; adding a short table or footnote list would improve traceability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important considerations for the evidential basis and scope of our reflective analysis. We address each major comment below and outline targeted revisions.

read point-by-point responses

Referee: The central claim that failures stem primarily from social and institutional factors rather than technical limitations is load-bearing for the paper's conclusion, yet it rests on narrative case studies without quantitative metrics, controlled comparisons to successful projects, or external validation. This weakens the causal attribution (see the three-failures discussion and the final paragraph).

Authors: We acknowledge that the central claims draw from interpretive case studies based on direct practitioner experience rather than quantitative metrics, controlled comparisons, or external validation. This approach is deliberate for a reflective paper that extracts lessons from two decades of infrastructure-building work, where social and institutional dynamics are often documented through narrative rather than experimental design. To address the concern, we will revise the three-failures discussion and concluding paragraph to explicitly frame the attributions as interpretive insights from experience, add a short methodological note on the limitations of case-based reflection, and avoid stronger causal language. revision: partial
Referee: The assertion that the author's experiences with Arabic are representative of challenges for other underserved languages is stated without supporting evidence or discussion of selection effects. A concrete test would be to compare the identified social barriers against documented cases for at least one additional language family.

Authors: The manuscript presents the lessons as emerging from our specific Arabic NLP experience and suggests potential relevance to other underserved languages without claiming full representativeness or providing systematic evidence. We agree that explicit discussion of selection effects would improve clarity. However, incorporating a detailed comparison to another language family would expand the paper beyond its intended scope as a focused practitioner reflection. We will revise the final paragraph and lessons section to include a brief qualification on selection effects and to present the findings as suggestive rather than definitive for other contexts. revision: partial

Circularity Check

0 steps flagged

No significant circularity in reflective narrative

full rationale

This paper is a reflective essay based on the author's personal experiences across Arabic NLP projects rather than a technical contribution with equations, derivations, or fitted parameters. The central claims about social, institutional, and epistemic challenges are presented as interpretive lessons from specific failures and observations, without reducing to self-citations or prior inputs by construction. No load-bearing steps match the enumerated circularity patterns; the argument is self-contained within its narrative genre and draws on direct experience without formal proof or parameter fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is a personal reflection on two decades of Arabic NLP projects and introduces no mathematical models, fitted parameters, background axioms, or newly postulated entities.

pith-pipeline@v0.9.0 · 5711 in / 1117 out tokens · 32808 ms · 2026-05-21T05:14:20.584561+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

[1]

Abdul-Mageed, M., Elmadany, A., and Nagoudi, E. M. B. (2021). ARBERT and MARBERT: Deep Bidirectional Transformers for Arabic. In Proceed- ings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL-IJCNLP 2021), pages 7088–7105. Ali Al-Athba, A. and Zaghou...

work page 2021
[2]

and Zaghouani, W

Al Heraki, H. and Zaghouani, W. (2025). Analyzing Digital Polarization on Hijab: A Dataset of Anno - tated YouTube Comments. In Proceedings of ICWSM 2025, pages 2350–2360. Bashiti, A., Aljabari, A., Hamoud, H. K., Biswas, M. R., Shalash, B. M., Jarrar, M., Zaraket, F., Mikros, G., Asgari, E., and Zaghouani, W. (2025). ImageEval 2025: The First Arabic Imag...

work page 2025
[3]

Birhane, A., Isaac, W., Prabhakaran, V., Diaz, M., Elish, M

In Proceedings of ArabicNLP 2025 Shared Tasks , pages 998–1002. Birhane, A., Isaac, W., Prabhakaran, V., Diaz, M., Elish, M. C., Gabriel, I., and Mohamed, S. (2022). Power to the People? Opportunities and Challenges for Par- ticipatory AI. In Proceedings of the 2nd ACM Con - ference on Equity and Access in Algorithms, Mecha- nisms, and Optimization (EAAMO...

work page 2025
[4]

(Eds.) (2022)

Bouamor, H., Al Khalifa, H., Bougares, F., Darwish, K., Rambow, O., Abdelali, A., Tomeh, N., Khalifa, S., and Zaghouani, W. (Eds.) (2022). Proceedings of the Seventh Arabic Natural Language Processing Work- shop. Association for Computational Linguistics. Bowman, S. R. and Dahl, G. E. (2021). What Will it Take to Fix Benchmarking in Natural Language Under...

work page 2022
[5]

and Jurafsky, D

Ethayarajh, K. and Jurafsky, D. (2020). Utility is in the Eye of the User: A Critique of NLP Leaderboards. In Proceedings of EMNLP 2020, pages 4846–4853. Fanelli, D. (2012). Negative Results Are Disappearing from Most Disciplines and Countries. Scientomet- rics, 90(3):891–904. Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé I...

work page 2020
[6]

Ibrahim, S., Biswas, M

In Working Notes of CLEF 2024, pages 276–286. Ibrahim, S., Biswas, M. R., Bessghaier, M., and Za - ghouani, W. (2025). MarsadLab at BAREC Shared Task

work page 2024
[7]

Maamouri, M., Bies, A., Kulick, S., Zaghouani, W., Graff, D., and Ciul, M

In Proceedings of ArabicNLP 2025 Shared Tasks, pages 274–279. Maamouri, M., Bies, A., Kulick, S., Zaghouani, W., Graff, D., and Ciul, M. (2010). From Speech to Trees: Applying Treebank Annotation to Arabic Broadcast News. In Proceedings of LREC

work page 2025
[8]

Mohit, B., Rozovskaya, A., Habash, N., Zaghouani, W., and Obeid, O. (2014). The First QALB Shared Task on Automatic Text Correction for Arabic. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 39–47. Nakov, P., Barrón -Cedeño, A., Elsayed, T., Suwaileh, R., Màrquez, L., Zaghouani, W., Atanasova, P., Kyuchukov, ...

work page 2014
[9]

Plank, B. (2022). The “Problem” of Human Label Vari- ation: On Ground Truth in Data, Modeling and Eval- uation. In Proceedings of EMNLP

work page 2022
[10]

Rangel, F., Rosso, P., Charfi, A., Zaghouani, W., Ghanem, B., and Sánchez -Junquera, J. (2019). Overview of the Track on Author Profiling and De - ception Detection in Arabic. In Working Notes of FIRE 2019, pages 70–83. CEUR Workshop Proceed- ings. Rangel, F., Rosso, P., Zaghouani, W., and Charfi, A. (2020). Fine-grained Analysis of Language Varieties and...

work page 2019
[11]

Sap, M., Card, D., Gabriel, S., Choi, Y., and Smith, N. A. (2019). The Risk of Racial Bias in Hate Speech Detection. In Proceedings of ACL 2019, pages 1668–

work page 2019
[12]

Sharqawi, E. A. and Zaghouani, W. (2026). Ara- HopeCorpus: Annotation Guidelines and Dataset for Hope Speech in Arabic Social Media Crisis Dis- course. In Proceedings of LREC

work page 2026
[13]

and Zaghouani, W

Shestakov, A. and Zaghouani, W. (2024). Analyzing Conflict Through Data: A Dataset on the Digital Framing of Sheikh Jarrah Evictions. In Proceed- ings of the Second Workshop on NLP for Political Sciences, LREC-COLING 2024, pages 55–67. Shurafa, C., Darwish, K., and Zaghouani, W. (2020). Political Framing: US COVID -19 Blame Game. In Social Informatics (SocInfo

work page 2024
[14]

Springer

, LNCS 12467. Springer. Zaghouani, W. (2010). Arabic Treebank Part 1 Version 4.1. LDC Catalog No. LDC2010T13. Linguistic Data Consortium. Zaghouani, W., Pouliquen, B., Ebrahim, M., and Stein- berger, R. (2010). Adapting a Resource-Light Highly Multilingual Named Entity Recognition System to Arabic. In Proceedings of LREC

work page 2010
[15]

Zaghouani, W., Diab, M., Mansouri, A., Pradhan, S., and Palmer, M. (2010). The Revised Arabic Prop - Bank. In Proceedings of the 4th Linguistic Annota - tion Workshop, ACL

work page 2010
[16]

Zaghouani, W. (2012). RENAR: A Rule-Based Arabic Named Entity Recognition System. ACM Transac- tions on Asian Language Information Processing , 11(1), Article

work page 2012
[17]

Zaghouani, W., Mohit, B., Habash, N., Obeid, O., Tomeh, N., Rozovskaya, A., Farra, N., Alkuhlani, S., and Oflazer, K. (2014). Large Scale Arabic Error Annotation: Guidelines and Framework. In Proceed- ings of LREC 2014, pages 2362–2369. Zaghouani, W., Habash, N., Bouamor, H., Rozovskaya, A., Mohit, B., Heider, A., and Oflazer, K. (2015). Correction Annota...

work page 2014
[18]

Zaghouani, W., Hawwari, A., Diab, M., O’Gorman, T., and Badran, A. (2016). AMPN: A Semantic Re - source for Arabic Morphological Patterns. Interna- tional Journal of Speech Technology, 19(2):281–288. Zaghouani, W. and Charfi, A. (2018). AraP-Tweet: A Large Multi-Dialect Twitter Corpus for Gender, Age and Language Variety Identification. In Proceedings of LREC

work page 2016
[19]

Zaghouani, W. (2018). A Large -Scale Social Media Corpus for the Detection of Youth Depression. Pro- cedia Computer Science, 142:347–351. Zaghouani, W., Mubarak, H., and Biswas, M. R. (2024). So Hateful! Building a Multi -Label Hate Speech Annotated Arabic Dataset. In Proceedings of LREC- COLING 2024, pages 15044–15055. Zaghouani, W. and Biswas, M. R. (20...

work page 2018
[20]

R., and Ibrahim, S

Zaghouani, W., Bessghaier, M., Biswas, M. R., and Ibrahim, S. A. (2026). Audience Engagement with Arabic Women’s Social Empowerment and Wellbe- ing: A Decadal Corpus. In Proceedings of LREC

work page 2026
[21]

(Eds.) (2020)

Zitouni, I., Abdul-Mageed, M., Bouamor, H., Bougares, F., El-Haj, M., Tomeh, N., and Zaghouani, W. (Eds.) (2020). Proceedings of the Fifth Arabic Natural Lan- guage Processing Workshop . Association for Com- putational Linguistics

work page 2020

[1] [1]

Abdul-Mageed, M., Elmadany, A., and Nagoudi, E. M. B. (2021). ARBERT and MARBERT: Deep Bidirectional Transformers for Arabic. In Proceed- ings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL-IJCNLP 2021), pages 7088–7105. Ali Al-Athba, A. and Zaghou...

work page 2021

[2] [2]

and Zaghouani, W

Al Heraki, H. and Zaghouani, W. (2025). Analyzing Digital Polarization on Hijab: A Dataset of Anno - tated YouTube Comments. In Proceedings of ICWSM 2025, pages 2350–2360. Bashiti, A., Aljabari, A., Hamoud, H. K., Biswas, M. R., Shalash, B. M., Jarrar, M., Zaraket, F., Mikros, G., Asgari, E., and Zaghouani, W. (2025). ImageEval 2025: The First Arabic Imag...

work page 2025

[3] [3]

Birhane, A., Isaac, W., Prabhakaran, V., Diaz, M., Elish, M

In Proceedings of ArabicNLP 2025 Shared Tasks , pages 998–1002. Birhane, A., Isaac, W., Prabhakaran, V., Diaz, M., Elish, M. C., Gabriel, I., and Mohamed, S. (2022). Power to the People? Opportunities and Challenges for Par- ticipatory AI. In Proceedings of the 2nd ACM Con - ference on Equity and Access in Algorithms, Mecha- nisms, and Optimization (EAAMO...

work page 2025

[4] [4]

(Eds.) (2022)

Bouamor, H., Al Khalifa, H., Bougares, F., Darwish, K., Rambow, O., Abdelali, A., Tomeh, N., Khalifa, S., and Zaghouani, W. (Eds.) (2022). Proceedings of the Seventh Arabic Natural Language Processing Work- shop. Association for Computational Linguistics. Bowman, S. R. and Dahl, G. E. (2021). What Will it Take to Fix Benchmarking in Natural Language Under...

work page 2022

[5] [5]

and Jurafsky, D

Ethayarajh, K. and Jurafsky, D. (2020). Utility is in the Eye of the User: A Critique of NLP Leaderboards. In Proceedings of EMNLP 2020, pages 4846–4853. Fanelli, D. (2012). Negative Results Are Disappearing from Most Disciplines and Countries. Scientomet- rics, 90(3):891–904. Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé I...

work page 2020

[6] [6]

Ibrahim, S., Biswas, M

In Working Notes of CLEF 2024, pages 276–286. Ibrahim, S., Biswas, M. R., Bessghaier, M., and Za - ghouani, W. (2025). MarsadLab at BAREC Shared Task

work page 2024

[7] [7]

Maamouri, M., Bies, A., Kulick, S., Zaghouani, W., Graff, D., and Ciul, M

In Proceedings of ArabicNLP 2025 Shared Tasks, pages 274–279. Maamouri, M., Bies, A., Kulick, S., Zaghouani, W., Graff, D., and Ciul, M. (2010). From Speech to Trees: Applying Treebank Annotation to Arabic Broadcast News. In Proceedings of LREC

work page 2025

[8] [8]

Mohit, B., Rozovskaya, A., Habash, N., Zaghouani, W., and Obeid, O. (2014). The First QALB Shared Task on Automatic Text Correction for Arabic. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 39–47. Nakov, P., Barrón -Cedeño, A., Elsayed, T., Suwaileh, R., Màrquez, L., Zaghouani, W., Atanasova, P., Kyuchukov, ...

work page 2014

[9] [9]

Plank, B. (2022). The “Problem” of Human Label Vari- ation: On Ground Truth in Data, Modeling and Eval- uation. In Proceedings of EMNLP

work page 2022

[10] [10]

Rangel, F., Rosso, P., Charfi, A., Zaghouani, W., Ghanem, B., and Sánchez -Junquera, J. (2019). Overview of the Track on Author Profiling and De - ception Detection in Arabic. In Working Notes of FIRE 2019, pages 70–83. CEUR Workshop Proceed- ings. Rangel, F., Rosso, P., Zaghouani, W., and Charfi, A. (2020). Fine-grained Analysis of Language Varieties and...

work page 2019

[11] [11]

Sap, M., Card, D., Gabriel, S., Choi, Y., and Smith, N. A. (2019). The Risk of Racial Bias in Hate Speech Detection. In Proceedings of ACL 2019, pages 1668–

work page 2019

[12] [12]

Sharqawi, E. A. and Zaghouani, W. (2026). Ara- HopeCorpus: Annotation Guidelines and Dataset for Hope Speech in Arabic Social Media Crisis Dis- course. In Proceedings of LREC

work page 2026

[13] [13]

and Zaghouani, W

Shestakov, A. and Zaghouani, W. (2024). Analyzing Conflict Through Data: A Dataset on the Digital Framing of Sheikh Jarrah Evictions. In Proceed- ings of the Second Workshop on NLP for Political Sciences, LREC-COLING 2024, pages 55–67. Shurafa, C., Darwish, K., and Zaghouani, W. (2020). Political Framing: US COVID -19 Blame Game. In Social Informatics (SocInfo

work page 2024

[14] [14]

Springer

, LNCS 12467. Springer. Zaghouani, W. (2010). Arabic Treebank Part 1 Version 4.1. LDC Catalog No. LDC2010T13. Linguistic Data Consortium. Zaghouani, W., Pouliquen, B., Ebrahim, M., and Stein- berger, R. (2010). Adapting a Resource-Light Highly Multilingual Named Entity Recognition System to Arabic. In Proceedings of LREC

work page 2010

[15] [15]

Zaghouani, W., Diab, M., Mansouri, A., Pradhan, S., and Palmer, M. (2010). The Revised Arabic Prop - Bank. In Proceedings of the 4th Linguistic Annota - tion Workshop, ACL

work page 2010

[16] [16]

Zaghouani, W. (2012). RENAR: A Rule-Based Arabic Named Entity Recognition System. ACM Transac- tions on Asian Language Information Processing , 11(1), Article

work page 2012

[17] [17]

Zaghouani, W., Mohit, B., Habash, N., Obeid, O., Tomeh, N., Rozovskaya, A., Farra, N., Alkuhlani, S., and Oflazer, K. (2014). Large Scale Arabic Error Annotation: Guidelines and Framework. In Proceed- ings of LREC 2014, pages 2362–2369. Zaghouani, W., Habash, N., Bouamor, H., Rozovskaya, A., Mohit, B., Heider, A., and Oflazer, K. (2015). Correction Annota...

work page 2014

[18] [18]

Zaghouani, W., Hawwari, A., Diab, M., O’Gorman, T., and Badran, A. (2016). AMPN: A Semantic Re - source for Arabic Morphological Patterns. Interna- tional Journal of Speech Technology, 19(2):281–288. Zaghouani, W. and Charfi, A. (2018). AraP-Tweet: A Large Multi-Dialect Twitter Corpus for Gender, Age and Language Variety Identification. In Proceedings of LREC

work page 2016

[19] [19]

Zaghouani, W. (2018). A Large -Scale Social Media Corpus for the Detection of Youth Depression. Pro- cedia Computer Science, 142:347–351. Zaghouani, W., Mubarak, H., and Biswas, M. R. (2024). So Hateful! Building a Multi -Label Hate Speech Annotated Arabic Dataset. In Proceedings of LREC- COLING 2024, pages 15044–15055. Zaghouani, W. and Biswas, M. R. (20...

work page 2018

[20] [20]

R., and Ibrahim, S

Zaghouani, W., Bessghaier, M., Biswas, M. R., and Ibrahim, S. A. (2026). Audience Engagement with Arabic Women’s Social Empowerment and Wellbe- ing: A Decadal Corpus. In Proceedings of LREC

work page 2026

[21] [21]

(Eds.) (2020)

Zitouni, I., Abdul-Mageed, M., Bouamor, H., Bougares, F., El-Haj, M., Tomeh, N., and Zaghouani, W. (Eds.) (2020). Proceedings of the Fifth Arabic Natural Lan- guage Processing Workshop . Association for Com- putational Linguistics

work page 2020