Second Call for Participation: ImageEval 2026: Cultural Grounding in Arabic Multimodal Generation and Understanding
We invite researchers and practitioners to participate in ImageEval 2026: Cultural Grounding in Arabic Multimodal Generation and Understanding, a shared task organized as part of ArabicNLP 2026, co-located with EMNLP 2026.
ImageEval 2026 focuses on Arabic language varieties and culturally specific visual content from the MENA region. The task suite evaluates visual question answering, image-grounded hallucination detection, and cultural accuracy evaluation for text-to-image generation, with dedicated tracks for participants to build, evaluate, and compare multimodal systems.
Tracks
- Task 1: Culturally Grounded Arabic Visual Question Answering and Hallucination Detection
- Task 2: Cultural Accuracy Evaluation for Text-to-Image Generation
Participants are invited to develop systems, submit predictions, and write system description papers for the ArabicNLP 2026 proceedings.
Registration:
- Spoken VQA (1a) English: https://www.codabench.org/competitions/17049/
- Spoken VQA (1a) MSA: https://www.codabench.org/competitions/17048/
- Hallucination (1b) English: https://www.codabench.org/competitions/17051/
- Hallucination (1b) MSA: https://www.codabench.org/competitions/17050/
- CRAI-Bench: https://www.codabench.org/competitions/16944/
Links
- Website: https://imageeval2026.github.io/
- Data and code: https://github.com/ImageEval2026/ImageEval2026-tasks
We encourage participation from the NLP, multimodal AI, computer vision, online safety, LLM safety, and Arabic language technology communities.