Dataset

Dataset

For the MMCultureQA SemEval 2027 shared task, we are curating language-specific datasets for all participating tracks.

The datasets for English, Modern Standard Arabic (MSA), Egyptian Arabic, and Levantine Arabic are derived from OASIS, where each image is paired with culturally grounded questions and open-ended answers in both spoken and written form.

For the remaining languages, we follow the same EverydayMMQA data-development pipeline used to curate OASIS. The data covers 9 broad cultural topic categories and 31 sub-categories, including culturally grounded concepts related to places, food, traditions, everyday objects, and social practices.

Language-specific datasets will be released gradually throughout the shared-task preparation period and before the evaluation phase. Participants should check this page, the timeline, and the Hugging Face repository for the latest releases.

Below is an example of an image and its associated multimodal QA record.

Sample record image · audio · text

question_en

What is the name of the building shown in the image, and what inspired its design?

question_ar

ما اسم المبنى الموضح في الصورة، وما الذي ألهم تصميمه؟

question_audio

the same question, spoken

answer_en

The building is the National Museum of Qatar, and its design is inspired by desert rose formations.

answer_ar

المبنى هو متحف قطر الوطني، وتصميمه مستوحى من تشكيلات الورود الصحراوية.

The same question as audio and as text, in English and Arabic varieties; the target is a short open-ended answer. Photo: Msarg77, CC BY-SA 4.0.

Getting the data

The shared-task datasets are hosted on Hugging Face.

Data for individual languages will be released progressively, rather than all at once. Please check the timeline and the Hugging Face repository regularly for the availability of each language-specific dataset.

All language tracks are expected to have the required data available before the corresponding evaluation phase.

License

The shared-task dataset is planned for release under the CC BY-NC-SA 4.0 license: free for non-commercial research use, with attribution and share-alike.

Citation

If you use the MMCultureQA shared-task data, please cite the relevant dataset papers listed below. We will continue to add citations for newly released language-specific datasets as they become available.

@article{alam2025everydaymmqa,
  title = {{OASIS}: A Multilingual and Multimodal Dataset for Culturally Grounded Spoken Visual QA},
  author = {Alam, Firoj and Shahroor, Ali Ezzat and Hasan, Md. Arid and Ali, Zien Sheikh and Bhatti, Hunzalah Hassan and Kmainasi, Mohamed Bayan and Chowdhury, Shammur Absar and Mousi, Basel and Dalvi, Fahim and Durrani, Nadir and Milic-Frayling, Natasa},
  journal = {arXiv preprint arXiv:2510.06371},
  year = {2025}
}