Frequently Asked Questions

No. You may choose to participate in one or more subtasks.

No. You can choose one or more languages.

Yes. You may work on multiple languages at the same time.

We will announce the release of each dataset as it becomes available. Currently, datasets for 9 languages have been released, and the remaining ones will follow soon.

To be included in the final team rankings, participants must submit a system description paper detailing their approach and methodology. This ensures transparency and scientific integrity.

No. A system description paper is mandatory for inclusion in the final ranking.

We will provide an online tutorial and resources to guide you through the process.

Each participant can be part of only one team. If you believe there are special circumstances that require you to join more than one team, please email us before the evaluation period begins.

You may work either as part of a team or independently — but not both.

To add members to your team on Codabench:

• Click your account name in the upper-right corner and select your organization.
• On the organization page, click “Edit Organization” in the upper-left corner.
• Scroll to the bottom and click the green “Invite Members” button to add teammates.

• Dev set: Released at the start of the evaluation phase.
• Test set(s): Released after the competition ends.

Yes.

No. You may use any type or size of model — open-source or closed-source. You may also generate or use synthetic data for augmentation.

Yes, but you must properly cite all external datasets in your system description paper.

• Data sources include news websites, Reddit, blogs, Bluesky, and regional forums, covering topics such as elections, conflicts, gender rights, and migration.
• Each language dataset contains approximately 3,000–5,000 annotated instances. Annotation platforms include Label Studio, Prolific, Potato, and Mechanical Turk.

No. All data was annotated by at least three native speakers per instance — no LLMs were used.

You can access all dataset download links through this document: Google Drive Dataset Access Guide .

The starter pack is an example to help you get started with the data. You may use it as a reference or build your own approach from scratch.

The public data contain all subtasks, so you only need to download them once even if you plan to participate in multiple subtasks.

The starter pack includes a sample notebook that trains a small BERT model for demonstration. You are free to use your own code and models instead.

No. Publication does not require attendance. If you do not attend the workshop, you do not need to pay any fees. If you wish to attend, standard registration fees apply.

Yes. We strongly encourage all teams to submit a paper. Negative results are equally valuable and contribute to advancing research on polarization.

Codabench may occasionally experience delays in processing submissions. Please refresh the page after a few minutes. If the issue persists, try resubmitting your file.

Click on your submission in the leaderboard or the “My Submissions” tab. You can view the error log under the “Output” section to see detailed feedback about the failure.

This usually happens if your submission format doesn’t match the expected structure. Verify that your file follows the required format described on the evaluation page, then re-submit.

You cannot delete a submission, but you can simply re-submit a new file. Only the latest valid submission will be considered for the leaderboard during evaluation.

Ensure you are logged in and registered for the competition. The “Submit” button appears under the corresponding subtask page once registration is complete.

Submissions are only visible within your team. Make sure all members are added under the same organization on Codabench using the “Invite Members” option.