Source dialect text
Participants will receive text written in a specified Arabic dialect. The source dialects and input format are to be confirmed.
AbjadNLP 2027 · Shared task
Cross-dialect translation for Arabic varieties
A proposed shared task exploring translation between Arabic dialects, with an additional dialect planned for the 2027 edition. Language pairs, datasets and evaluation arrangements will be confirmed by the organisers.
Illustrative phrases only, not examples from the dataset. Dialect pairs are to be confirmed.
About the shared task
Arabic dialects vary in vocabulary, spelling, morphology and expression. This shared task will investigate how well systems can transfer meaning between dialects while producing natural, appropriate target-dialect text.
Participants will receive text written in a specified Arabic dialect. The source dialects and input format are to be confirmed.
Systems will generate an equivalent text in a designated target Arabic dialect, aiming to preserve the source meaning.
The proposal extends an earlier cross-dialect translation task by adding one further Arabic dialect. The full set of dialect pairs is TBC.
Automatic evaluation measures and any human assessment of meaning preservation, fluency and dialect appropriateness will be announced by the organisers.
Parallel data sources, permitted training resources, dialect pairings, data licences, train/development/test splits and official evaluation metrics have not yet been confirmed. These details will be published before participation opens.
Taking part
Registration, data access and evaluation instructions will be announced when the shared-task arrangements are confirmed.
This is a preliminary description. Refer to the workshop website for announcements and applicable policies.
Visit the AbjadNLP 2027 workshop page →The team
The previous edition's organising team is shown below; the 2027 team is subject to confirmation.