Afri-MCQA: Multimodal Cultural Question Answering for African Languages

Paper

Overview

Afri-MCQA is the first multilingual cultural question-answering benchmark covering 8k Q&A pairs across 16 African languages from 13 countries. The benchmark offers parallel English-African language Q&A pairs across text and speech modalities, entirely created by native speakers.

Supported Tasks

• Visual Question Answering (VQA): Multiple-choice and open-ended QA grounded in culturally relevant images
• Visual Audio Question Answering: Speech-based QA grounded in culturally relevant images in native African languages and African-accented English
• Language Identification (LID): Identifying which of the 15 languages is spoken
• Automatic Speech Recognition (ASR): Transcribing spoken African language audio

Languages

Language - Country	Language Family	Region
Akan/Twi - Ghana	Niger-Congo / Volta-Niger	West
Amharic - Ethiopia	Afro-Asiatic / Ethio-Semitic	East
Chichewa - Malawi	Niger-Congo / Bantu	South & East
Hausa - Nigeria	Afro-Asiatic / Chadic	West
Igbo - Nigeria	Niger-Congo / Volta-Niger	West
Kikuyu - Kenya	Niger-Congo / Bantu	East
Kinyarwanda - Rwanda	Niger-Congo / Bantu	East
Lingala - DRC	Niger–Congo/ Bantu	Central Africa
Luganda - Uganda	Niger-Congo / Bantu	East
Oromo - Ethiopia	Afro-Asiatic / Cushitic	East
Setswana - Botswana	Niger-Congo / Bantu	South
Somali - Somalia	Afro-Asiatic / Cushitic	East
Tigrinya - Eritrea	Afro-Asiatic / Ethio-Semitic	East
Yoruba - Nigeria	Niger-Congo / Volta-Niger	West
Sesotho - Lesotho	Niger-Congo / Bantu	South
Zulu - South Africa	Niger-Congo / Bantu	South

Total speaker population: ~412.6 million

Dataset Structure

Each sample includes:

• image: Culturally relevant image
• question_english / question_native: Question in both languages
• options_english / options_native: Four multiple-choice options
• answer: Correct answer
• audio_question_english / audio_question_native: Audio recordings
• category: One of 10 cultural categories
• country / language: Origin metadata

Cultural Categories

🏛️ Geography & Landmarks | 👤 Public Figures & Pop Culture | 🍲 Cooking & Food | 👕 Objects & Clothing |🎭 Traditions & History | 🏢 Brands & Companies | 🌿 Plants & Animals | 👨‍👩‍👧 People & Everyday Life | 🚗 Vehicles & Transportation | ⚽ Sports & Recreation

Licensing

This dataset is released under CC-BY-NC-4.0.

Usage

from datasets import load_dataset

# Per language
amharic_dev = load_dataset("Atnafu/Afri-MCQA", "Amharic_dev", split="dev")
amharic_test = load_dataset("Atnafu/Afri-MCQA", "Amharic_test", split="test")

# All languages grouped
all_dev = load_dataset("Atnafu/Afri-MCQA", "all_dev", split="dev")
all_test = load_dataset("Atnafu/Afri-MCQA", "all_test", split="test")

Citation

 @inproceedings{tonja-etal-2026-afri,
    title = "Afri-{MCQA}: Multimodal Cultural Question Answering for {A}frican Languages",
    author = "Tonja, Atnafu Lambebo  and
      Anand, Srija  and
      Villa-Cueva, Emilio  and
      Azime, Israel Abebe  and
      Alabi, Jesujoba Oluwadara  and
      Mohamed, Muhidin A.  and
      Yadeta, Debela Desalegn  and
      Abadi, Negasi Haile  and
      Oppong, Abigail  and
      Obiefuna, Nnaemeka Casmir  and
      Abdulmumin, Idris  and
      Etori, Naome A  and
      Wairagala, Eric Peter  and
      Tshinu, Kanda Patrick  and
      Emmanuel, Imanigirimbabazi  and
      Malema, Gabofetswe  and
      Aji, Alham Fikri  and
      Adelani, David Ifeoluwa  and
      Solorio, Thamar",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.acl-long.1869/",
    doi = "10.18653/v1/2026.acl-long.1869",
    pages = "40249--40282",
    ISBN = "979-8-89176-390-6",
    abstract = "Africa is home to over one-third of the world{'}s languages, yet remains severely underrepresented in multimodal AI research. We introduce Afri-MCQA, the first Multilingual Cultural Question-Answering benchmark containing 7.5k Q A pairs across 15 African languages from 12 countries. The benchmark offers parallel text and speech modalities and was entirely created by native speakers. We find that models show poor performance across evaluated cultures, with near-zero accuracy on open-ended VQA when queried through native language or speech. To test linguistic competence, we include control experiments meant to assess this specific aspect separate from cultural knowledge, and we observe significant performance gaps between native languages and English for both text and speech. These findings underscore the pressing need for speech-first approaches, culturally grounded pretraining, and cross-lingual cultural transfer. We release Afri-MCQA to support more inclusive multimodal AI development."
}