Frequently Asked Questions

Anyone can participate, either independently or as part of a team.

Participation in the shared task is free. Conference registration is only needed if your team presents its paper or attends the conference; publication of an accepted paper does not require registration.

Sign up on the relevant Codabench competition and get the data from the Hugging Face dataset. Join our Slack channel for announcements, deadlines, and task-specific updates.

Yes. Teams may participate in Task 1, Task 2, or both.

This depends on the task policy. If this is relevant to your team, contact the organizers before submitting runs.

For Task 1, any model and any pipeline are welcome, cascaded or end to end, and both open and closed models are allowed (we especially encourage open models for reproducibility and low-resource settings). Please take part in good faith: run each track in its own language from end to end rather than reusing another track’s predictions. See the Task 1 page for the full participation notes.

The evaluation period is the official competition window. Organizers release evaluation data at the beginning of this period, and system outputs are due before the deadline.

Submission limits are task-specific. Check the task page and platform instructions for the final limits.

Task 1 data is on Hugging Face. Task 2 data is linked from the Task 2 page.

Yes. Teams that participate and request task data are expected to submit a system description paper. Papers are due August 19, 2026 through OpenReview, and every team must also fill in the system description form by August 15, 2026.

Requesting the data commits the team to submitting a paper. Abandoning the task after receiving data may affect future participation.

ArabicNLP ran a webinar on writing a strong system description paper, presented by Dr Usman Naseem: watch the recording.

System papers should follow the ArabicNLP 2026 short-paper format. Papers must use the official EMNLP/ACL LaTeX or Word templates and may be up to 4 pages, excluding unlimited references. Templates are available from acl-org/acl-style-files.

Use this format:

<Team Name> at ImageEval 2026 Shared Tasks: <Your Contribution>

Example:

Scholarly at ImageEval 2026 Shared Tasks: LLMs for Arabic Visual Question Answering

Yes. Each team is required to cite the shared task overview paper in its system paper.

Submissions that exceed length requirements may be desk rejected.

No. Shared task papers, including system papers and overview papers, are not anonymous.

Yes. Papers are submitted through the ImageEval shared task venue on OpenReview. Create your account early, as verification can take a few days.

Yes. Shared task teams are expected to serve as reviewers, and reviews will be checked by a member of the organizing team.

Presenting your paper, in person or virtually, is encouraged. Registration is not required for an accepted paper to be published in the ACL Anthology; however, to present the paper or attend the conference, at least one author must register.

Yes. Authors may submit papers to arXiv at any time. Since shared task papers are not anonymous, this does not interfere with the review process.

No additions, revisions, or name changes are allowed after the review period for the camera-ready version unless explicitly permitted by the organizers.

Discuss ethical considerations, including potential misuse, clearly in the system paper.

Code and data release is highly encouraged. Include the URL in the system paper when possible to support reproducibility.

• For task-specific questions, consult the task page and contact the task organizers.
• For workshop logistics, contact the shared task organizers.
• For OpenReview technical issues, contact OpenReview support.