Registration is open for TariQA 2027. Register your team
TariQA 2027 · AbjadNLP Workshop

Paper Submission Guidelines

How to write and submit your system description paper. Papers are short, non-anonymous, reviewed by fellow participants, and published in the AbjadNLP 2027 proceedings.

1. Essentials

WhoEvery team with at least one official run in the evaluation phase.
WhatA system description paper: how your system works, how it was built, how it performed, what you learned.
LengthUp to 4 pages of main content. References and appendices are unlimited.
WhereOpenReview.
WhenSee the timeline. Papers submitted after the deadline cannot be included in the proceedings.
AnonymityPapers are not anonymous. You may post them on arXiv at any time.

2. Title

Use the pattern <Team Name> at TariQA 2027: <Your Contribution>, for example:

  • Team 1 at TariQA 2027: Retrieval-Augmented Answering for Darija Traffic Questions
  • Team 2 at TariQA 2027: Does Fine-tuning Beat Prompting on Moroccan Traffic Law?

The team name in the title must match the team name on Codabench.

3. Recommended structure

  1. Abstract A short summary of the system and the main result.
  2. Introduction Why the task matters, which language varieties you targeted, a citation of the task overview paper, your main strategy and key findings. A link to your code is strongly recommended here.
  3. Background The task setup with an input/output example, the dataset (varieties, sources, size), and the related work that shaped your approach.
  4. System overview Algorithms, models and design decisions; external resources (models, corpora, legal texts, tools) and how they were used; how you handled Darija, code-mixing and article-grounded questions; equations or pseudo-code for anything novel. If you submitted several runs or configurations, make the differences explicit.
  5. Experimental setup How you used the train, development and test splits; preprocessing; hyper-parameters; prompts (verbatim, in the appendix if long); software with versions and URLs; evaluation metrics.
  6. Results The official test score and rank, clearly separated from post-evaluation experiments; development-set results, ablations and comparisons; a breakdown by language variety and question type; an error analysis with examples. State which split every number comes from.
  7. Conclusion Summary, limitations, and what you would do next.
  8. Ethics and limitations A short paragraph on possible misuse, in particular the risk of wrong legal advice to drivers, and on the limits of your system.
  9. Acknowledgments Optional.
  10. Appendix Low-level details needed for replication: full prompts, retrieval settings, additional tables.

4. Formatting

  • Use the official ACL style files (LaTeX or Word), exactly as distributed. Do not modify margins, fonts or spacing, and do not use templates from other venues.
  • Main content: up to 4 pages. References: unlimited. Appendices: unlimited, but reviewers are not required to read them, so everything needed to understand and judge the system belongs in the main text.
  • Submissions that exceed the page limit or do not conform to the style are returned without review.
  • Write in English. Examples from the data may be given in Arabic script with a translation or gloss.
  • Cite the TariQA overview paper (see below) and any dataset, model or tool you used.

5. Guiding principles

  • Replicability. A reader should be able to rebuild your system from the paper and the appendix. Release code and models when you can, and give the URL in the paper.
  • Analysis over ranking. A well-analysed system that did not top the leaderboard is more useful to the community than a bare score. Report multiple runs, ablations, and failures.
  • Do not repeat the task paper. Describe the task in a few sentences and cite the overview paper for the rest.
  • Be honest about the official run. Post-evaluation improvements are welcome, but must be labelled as such and kept apart from the official result.

6. Review and camera-ready

  • Every author needs an OpenReview account.
  • Each team reviews two other system papers. Reviews are short, constructive and checked by an organizer. Not reviewing may lead to the team's own paper being withdrawn.
  • Acceptance is based on clarity, completeness and replicability, not on leaderboard position.
  • Camera-ready versions may only change what the reviewers and organizers asked for. No new experiments or claims after the review period.

7. Citing the task

The overview paper will be available before the paper deadline. Until then, cite the task as follows and update the entry in the camera-ready version:

@inproceedings{tariqqa2027,
  title     = {TariQA 2027: Shared Task on Moroccan Traffic Law Question Answering
               in Modern Standard Arabic and Darija},
  author    = {ORGANIZERS, TBA},
  booktitle = {Proceedings of the AbjadNLP 2027 Workshop},
  year      = {2027}
}

8. Checklist before you submit

  • Title follows <Team> at TariQA 2027: <Contribution> and matches the Codabench team name.
  • Official ACL style, 4 pages of main content, no modified template.
  • The official test result is reported and separated from post-evaluation results.
  • Every external resource (models, APIs, legal texts, extra data) is named with a version.
  • Per-variety results and an error analysis are included.
  • The overview paper is cited; a code URL is given if available.
  • Ethics and limitations paragraph is present.
  • All authors have OpenReview accounts.
Submit on OpenReview Participation guidelines Back to the task page