Subject: Call for Participation: AraSentEval 2026 Shared Task on Sentiment Analysis and Swapping in Arabic (Hosted with OSACT7 at LREC 2026)
We are delighted to invite you to participate in the AraSentEval 2026 Shared Task, hosted in conjunction with the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) at LREC 2026.
Sentiment analysis remains a critical application in NLP, yet Arabic presents unique challenges due to its rich morphology and dialectal variations. AraSentEval 2026 aims to push the boundaries of Arabic sentiment analysis by evaluating both understanding (classification) and generation (style transfer) in diverse contexts.
AraSentEval features two complementary subtasks designed to foster innovation in Arabic NLP:
A multi-class classification task to identify the sentiment (positive, negative, or neutral) of texts written in various Arabic dialects (Moroccan, Egyptian, Jordanian, and Saudi). This subtask focuses on robustness across lexical and syntactic diversity.
A challenging generative task where systems must rewrite an Arabic sentence to invert its sentiment polarity (e.g., positive to negative) while preserving its core meaning. This task requires deep semantic understanding and high-quality generation capabilities.
Participants are welcome to submit to one or both subtasks. The shared task will be managed via CodaLab.
For more details, participation guidelines, and data access, please visit our task website: https://ezzini.github.io/AraSentEval/
We look forward to your participation!
Best regards,
The AraSentEval 2026 Organizers