ImageEval 2026

Cultural Grounding in Arabic Multimodal Generation and Understanding

Paper Submission Guidelines

System description papers should explain what your team built, how it was trained and evaluated, and what the results reveal about the task. Keep the paper concise, reproducible, and analysis-focused.

Paper length Maximum 4 pages

References are unlimited.

Required template EMNLP 2026 / ACL style

LaTeX or Word templates.

Title format Team at ImageEval

<Team Name> at ImageEval Shared Task: <Your Contribution>

Key Principles

Replicability

Provide enough implementation detail for another researcher to reproduce the system.

Analysis

Emphasize results, error patterns, ablations, and design decisions rather than only rankings.

Clarity

Briefly describe the task setup, but do not duplicate the task overview paper.

Required Elements

  • Cite the task overview paper.
  • Follow the EMNLP/ACL templates exactly.
  • Use the required title format: <Team Name> at ImageEval Shared Task: <Your Contribution>.
  • Clearly state which task(s) and track(s) your team participated in.
  • Describe external data, tools, APIs, or models used beyond the released task data.
  • Distinguish official submitted results from post-submission experiments.

For popular algorithms, citation is usually sufficient. Full mathematical detail is only needed when it is central to your contribution. Move detailed hyperparameters and low-level implementation details to the appendix when space is limited.

1. Abstract

Briefly summarize the task, your approach, and the main results in a few sentences.

2. Introduction
  • Describe the task and why it matters.
  • Mention the language varieties and track(s) covered.
  • Cite the task overview paper.
  • Summarize your main system strategy.
  • Highlight key findings, ranking, and challenges discovered.
  • Include a code URL if available.
3. Background
  • Summarize the task setup, including input and output types.
  • Describe dataset details such as language, genre, and size.
  • State the tracks you participated in.
  • Cite related work and explain what is different about your system.
4. System Overview
  • Explain key algorithms and design decisions.
  • List resources used beyond the provided training data.
  • Describe how your system addressed task-specific challenges.
  • Include equations or pseudocode for novel methods.
  • Clearly distinguish multiple configurations or submitted runs.
5. Experimental Setup
  • Explain how train, development, and test splits were used.
  • Provide preprocessing details and hyperparameters needed for replication.
  • List external tools and libraries with versions and URLs.
  • Summarize the official evaluation metrics.
  • Put low-level details in the appendix if space is limited.
6. Results and Analysis
  • Report official metric performance and ranking.
  • Include ablations, comparisons, and design-decision analysis where possible.
  • Provide error analysis with representative examples.
  • Clearly mark which data split is used for each analysis.
  • Distinguish official results from post-submission results.
7. Conclusion

Summarize the system, limitations, results, and future work.

8. Acknowledgments

Thank contributors, grants, infrastructure providers, and reviewers where appropriate.

9. Appendix

Use the appendix for low-level replication details that are useful but not essential to the main paper.

Formatting

  • Use the official EMNLP 2026 / ACL style templates in LaTeX or Word.
  • Download templates from acl-org/acl-style-files.
  • Follow the general ACL conference formatting guidelines.
  • Do not modify style files or use templates from other conferences.

Submissions with non-conforming paper size, margins, or font size may be rejected without review.

Final Checklist

  • The paper is no longer than 4 content pages.
  • References are in the correct format.
  • The title follows the required ImageEval format.
  • The task overview paper is cited.
  • All external data and tools are documented.
  • Official and post-submission results are clearly separated.
  • Code or data URLs are included when available.