Sorani Kurdish review sentence
A restaurant-review sentence written in Sorani Kurdish using the Perso-Arabic script.
AbjadNLP 2027 · Proposed shared task
Kurdish ASTE 2027
Aspect Sentiment Triplet Extraction in Sorani Kurdish
Extracting aspect terms, opinion expressions and sentiment polarity from Sorani Kurdish restaurant reviews written in the Perso-Arabic script.
Language and task
LanguageSorani Kurdish · Perso-Arabic script
Expected output(aspect, opinion, polarity)
The triplet notation illustrates the output structure, not an example from the dataset.
About the shared task
Fine-grained sentiment extraction
Rather than assigning one sentiment label to an entire review, systems identify the aspect being discussed, the expression conveying an opinion about it and whether that opinion is positive or negative. A sentence may contain zero, one or several triplets.
Aspect–opinion–polarity triplets
A list of triplets containing an aspect string, opinion string and a polarity label of positive or negative. Aspect and opinion strings are copied from the input.
Supervised generation
Participants may update model parameters, including fine-tuning encoder–decoder models or large language models on the training data.
Few-shot and prompt-based
Participants may use prompts, in-context examples and retrieval over training data, without updating model parameters. Track results are reported separately.
Dataset and evaluation
The proposal uses the publicly available KurdABSA dataset, containing Sorani Kurdish restaurant reviews annotated for aspect-based sentiment analysis. JSON and CSV formats are proposed. Exact training, development and test sizes remain to be confirmed.
The proposed main metric is exact-match triplet F1: the aspect string, opinion string and polarity must all match the gold annotation. The submitted proposal specifies a generative formulation; any extension to other modelling approaches or a revised evaluation protocol is still to be confirmed.
Taking part
Participation details
The task description is based on the submitted proposal. Final participation rules and arrangements will be announced by the organising team.
Participants may enter either or both proposed tracks. Please consult the workshop website for confirmed shared-task announcements and policies.
Visit the AbjadNLP 2027 workshop page →- Language
- Sorani Kurdish
- Domain
- Restaurant reviews
- Dataset
- KurdABSA
- Data split sizes
- To be confirmed
- Primary metric
- Exact-match triplet F1 (proposed)
- Registration and submissions
- To be announced
- Shared-task dates
- To be announced
The team
Shared-task organisers
The proposal currently identifies one organiser. Further organisers and affiliations will be added once confirmed.
- Rania AzadProposer · Affiliation to be confirmed