Question and three options
A traffic-law question in MSA or Darija, accompanied by three plausible candidate answers.
AbjadNLP 2027 · Shared task
TariQA 2027
Moroccan Traffic Law Question Answering
Evaluate how well language models and other NLP systems answer multiple-choice questions about Moroccan traffic regulations in Modern Standard Arabic (MSA) and Moroccan Darija.
Language coverage
Modern Standard Arabicقانون السير
Moroccan Darijaالدارجة المغربية
Illustrative phrases, not examples taken from the dataset. TariQA takes its name from the Arabic word for road, طريق, and QA for question answering.
About the shared task
Answering questions about Moroccan traffic law
Each question is presented with three candidate answers. Systems must select the single correct answer. The benchmark tests factual knowledge and the ability to apply traffic rules to described situations, across MSA and Darija.
One correct answer
Choose the correct option from the three candidates, following the submission format published by the organisers.
Factual recall
Questions about a directly stated rule, such as a speed limit, road sign or required vehicle document.
Rule application
Questions that describe a driving scenario and require applying the relevant rule, for example right of way or overtaking.
Dataset and evaluation
The proposal describes approximately 5,000 multiple-choice questions drawn from publicly accessible Moroccan driving-theory preparation resources and questions derived from Morocco’s Law No. 52-05 on the Highway Code. The final dataset size is subject to human validation and deduplication. Training labels will be available, while test labels will remain hidden. Accuracy is the primary metric; the proposed scoring script also reports F1. Results may be analysed by language variety and question type.
The organisers plan a zero-shot open-weight model baseline and a Codabench leaderboard. Data access, licensing, baseline release and final evaluation instructions will be confirmed by the organising team.
Taking part
Participation details
The organising team has prepared a dedicated TariQA website. Consult it for the latest registration, data access, submission and schedule information.
- Languages
- MSA and Moroccan Darija
- Task format
- Three-option multiple-choice QA
- Dataset size
- Approximately 5,000 (final count TBC)
- Primary metric
- Accuracy (F1 also proposed)
- Evaluation platform
- Codabench (planned)
- Registration and dates
- See task website; 2027 details TBC
The team
Shared-task organisers
Organisers and affiliations as listed in the TariQA shared-task proposal.
- Jihad R’baitiInternational Artificial Intelligence Center of Morocco, Mohammed VI Polytechnic University, Morocco
- Chouaib El HachimiResearch Centre for Artificial Intelligence in Geomatics, The Hong Kong Polytechnic University, Hong Kong
- Youssef HmamoucheInternational Artificial Intelligence Center of Morocco, Mohammed VI Polytechnic University, Morocco
- Emmanuele ChersoniDepartment of Language Science and Technology; Division of Artificial Intelligence and the Humanities, The Hong Kong Polytechnic University, Hong Kong
- Saad EzziniKing Fahd University of Petroleum and Minerals, Saudi Arabia
- Amal El Fallah SeghrouchniMohammed VI Polytechnic University, Morocco; LIP6, CNRS, Sorbonne University, France