Participants in the ArGuard Shared Tasks are invited to submit system description papers to the Fourth Arabic Natural Language Processing Conference—ArabicNLP 2026, co-located with EMNLP 2026.
The purpose of the system description papers is to document the methods, experiments, findings, and lessons learned during participation in ArGuard. Papers are welcome regardless of a system’s final leaderboard position. In addition to high-performing systems, we encourage submissions discussing unsuccessful approaches, unexpected results, limitations, and detailed error analyses.
Accepted papers will be published in the ArabicNLP 2026 shared-task proceedings by the Association for Computational Linguistics and made available through the ACL Anthology.
Important Dates
All deadlines are at 11:59 p.m. AoE—Anywhere on Earth (UTC−12).
| Event | Date |
|---|---|
| System description paper submission | August 18, 2026 |
| Acceptance notification | August 30, 2026 |
| Camera-ready paper submission | September 10, 2026 |
| ArabicNLP 2026 | October 28–29, 2026 |
Who May Submit?
All teams that participated in one or more ArGuard subtasks are invited to submit a system description paper. Teams that participated in multiple subtasks are encouraged to describe all their systems in a single paper. However, this is a recommendation rather than a strict requirement.
Paper Title
The paper title should follow this format:
<Team Name> at ArGuard Shared Tasks: <Paper Title>
For example:
Scholarly at ArGuard Shared Tasks: LLM Agents for Harmful Content Detection
Please use the same team name as in your official task submission and on the leaderboard. To change your team name, specify the new name in the form used to submit information about your system description paper. See Information Required for the Overview Paper.
Recommended Paper Content
The paper should provide a clear and reproducible description of the submitted system. You can follow the structure below.
1. Abstract
- Briefly summarize the proposed system, the task(s) addressed, the main methods, and the key results.
2. Introduction
- Introduce the task and its motivation.
- Summarize the main components and contributions of the submitted system.
3. Data and Preprocessing
- Describe the ArGuard training and development data used.
- Explain any text, image, or prompt preprocessing.
- Report any data augmentation or synthetic-data generation methods.
- Describe any external datasets, tools, or resources used.
4. System Architecture and Training
- Describe the models, encoders, classifiers, prompting strategies, agents, or processing pipelines.
- Report the training objectives and optimization settings.
- Specify the fine-tuning method, such as full fine-tuning, LoRA, or another parameter-efficient approach.
- Provide model checkpoints and important hyperparameters.
- Report the software frameworks and computational resources used.
5. Experimental Results
- Report the official test-set results.
- Include development-set results where relevant.
- Compare submitted runs, model variants, or system configurations.
- Present ablation studies where available.
6. Error Analysis
- Discuss representative errors made by the system.
- Identify difficult labels, linguistic phenomena, or content categories.
- Examine challenges related to Arabic dialects, code-switching, sarcasm, cultural references, visual context, and implicit intent.
- Discuss differences between development-set and test-set performance, where applicable.
7. Conclusion
- Summarize the main findings and contributions.
- Highlight possible directions for future work.
Limitations
- Discuss the limitations of the proposed approach.
- Report methods or configurations that did not perform as expected.
- Provide recommendations for future research.
Ethical Considerations
- Discuss potential risks associated with harmful-content modelling.
- Address privacy, bias, safety, and responsible-use considerations.
- Explain how sensitive, harmful, or offensive examples were handled.
The released gold-labelled test sets may be used for detailed error analysis in the system description paper.
Example papers:
Paper Length
System description papers may include up to four pages of main content.
References and appendices may use additional pages. Authors should ensure that the main system description, methodology, experimental results, and central conclusions are contained within the four-page main-paper limit. Note that Limitations and Ethical considerations will be after conclusions and can be in 5th page before References.
Paper Format
Papers must use the official two-column ACL format.
Authors must use the official ACL style files and must not:
- modify the ACL style files;
- change the required margins, font sizes, spacing, or page dimensions;
- use templates prepared for another conference; or
- manually adjust the layout to exceed the page limit.
Submissions that do not comply with the required ACL formatting specifications may be rejected without review.
ACL Templates
- ACL style-files repository: github.com/acl-org/acl-style-files
- Overleaf template: Association for Computational Linguistics (ACL) conference template
Authors should follow the general ACL formatting requirements.
Author Information and Review
ArGuard system description papers are not submitted anonymously. Authors should include their names, affiliations, and contact information in the submitted paper.
The papers will be reviewed to ensure that they:
- provide an adequate description of the submitted system;
- report the experimental setup and results clearly;
- follow the required paper format;
- include appropriate analysis or discussion;
- cite the ArGuard overview and dataset papers; and
- meet the basic standards required for inclusion in the proceedings.
A high leaderboard rank is not required for paper acceptance. The quality and clarity of the system description, analysis, and scientific discussion will be considered during review.
Paper Submission
Papers must be submitted through the official ArabicNLP 2026 shared-task paper submission portal.
Submission portal
Submission link to be announced
Before submitting, please confirm that:
- the paper uses the official ACL template;
- the paper contains no more than four pages of main content;
- the title follows the required naming format;
- all authors and affiliations are included;
- all external data and resources are disclosed;
- the official ArGuard papers are cited; and
- the submitted PDF opens and displays correctly.
ACL PubCheck for Camera-ready
All camera-ready papers must be checked using ACL PubCheck before final submission.
ACL PubCheck automatically identifies common formatting problems involving margins, fonts, page dimensions, spacing, and other ACL publication requirements.
Authors are also encouraged to run PubCheck before the initial paper submission.
PubCheck Resources
- GitHub repository: github.com/acl-org/aclpubcheck
- Google Colab: Run ACL PubCheck in Google Colab
- Hugging Face interface: ACL PubCheck on Hugging Face Spaces
Please address all relevant PubCheck errors before submitting the camera-ready paper. Warnings should also be reviewed carefully.
Required Citations
System description papers must cite:
- the ArGuard shared-task overview paper; and
- the relevant ArGuard dataset paper or papers.
ArGuard Shared-Task Overview Paper
@inproceedings{alam-etal-2026-arguard,
title = {{ArGuard Shared Task}: Harmful Content Detection in
{A}rabic {M}emes and {LLM} {P}rompts},
author = {Alam, Firoj and
Biswas, Md. Rafiul and
Kmainasi, Mohamed Bayan and
Shahroor, Ali Ezzat and
Mubarak, Hamdy and
Mikros, Georgios and
Hasnat, Abul and
Zaghouani, Wajdi},
booktitle = {Proceedings of the Fourth Arabic Natural Language
Processing Conference: Shared Tasks},
month = oct,
year = {2026},
address = {Budapest, Hungary},
publisher = {Association for Computational Linguistics}
}
Dataset Papers
The BibTeX entries for the relevant dataset papers will be provided below. Task A dataset-paper BibTeX to be announced
Information Required for the Overview Paper
The ArGuard organizers will publish an overview paper summarizing the shared task, participating teams, submitted methods, and official results.
Each participating team must provide:
- the team name;
- participating subtasks;
- a short description of the system;
- models and external resources used;
- the official submission results;
- authors and affiliations; and
- the BibTeX entry for the system description paper.
Please submit this information using the following form:
ArGuard overview paper information form
System Paper BibTeX Template
Please replace the placeholders with your paper information.
@inproceedings{arguard-2026-team-name,
author = {Last-Name, First-Name and
Last-Name, First-Name},
title = {{Team Name} at {ArGuard Shared Tasks}:
Title of the Paper},
booktitle = {Proceedings of the Fourth Arabic Natural Language
Processing Conference: Shared Tasks},
address = {Budapest, Hungary},
month = oct,
year = {2026},
publisher = {Association for Computational Linguistics}
}
Use a short and unique citation key based on your team name. Please ensure that the author names and paper title exactly match the submitted paper.
Leaderboards
Task A: Multimodal Hateful Meme Detection
- Subtask A1—Binary Classification: CodaBench competition 16909
- Subtask A2—Fine-Grained Classification: CodaBench competition 16910
Task B: Textual Harmful Prompt Detection
- Subtask B1—Binary Classification: CodaBench competition 16652
- Subtask B2—Harm-Category Classification: CodaBench competition 16653
Test Sets and Gold Labels
TBA
Call for Reviewers
Researchers who have previously published at *ACL conferences and are interested in serving as reviewers for the shared-task proceedings are invited to complete the following form: https://forms.gle/GAnDvu2cjqCDDSBQ9
Contact
For questions about paper preparation, submission, results, or the overview-paper form, please contact the ArGuard organizers.
- ArGuard website: araieval.github.io/ArGuard2026
- Task repository: github.com/araieval/ArGuard-2026-tasks
- Slack workspace: Join the ArAIEval Slack workspace
- Email: araieval@googlegroups.com