Datasets:

aistudioug
/

UCCB-Suite

annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - en
  - lg
  - nyn
  - ach
  - teo
  - lgg
license: cc-by-nc-sa-4.0
multilinguality:
  - multilingual
pretty_name: Ugandan Cultural Context Benchmark (UCCB) Suite — Version One
task_categories:
  - question-answering
task_ids:
  - open-domain-qa
size_categories:
  - 1K<n<10K
source_datasets:
  - original

Ugandan Cultural Context Benchmark (UCCB) Suite — Version One

Dataset Summary

The Ugandan Cultural Context Benchmark (UCCB) Suite is the first comprehensive question-answering dataset designed to evaluate the cultural understanding and reasoning abilities of Large Language Models (LLMs) concerning Uganda’s multicultural and multilingual environment.

This Version One builds upon the original 1,039 question-answer pairs across 24 cultural domains, with enhancements in language diversity, contextual clarity, and category balancing. It emphasizes linguistic richness, factual accuracy, and cultural nuance.

Supported Tasks and Leaderboards

Primary Task:

Question Answering (Open-book or Closed-book)
Suitable for both extractive and generative QA

Secondary Tasks:

Knowledge probing
Fine-tuning culturally aware models
Bias detection and mitigation

Leaderboard:

A public leaderboard is planned using an LLM-as-a-Judge rubric with cultural nuance scoring.

Languages

Primary Language: Ugandan English
Multilingual Elements: Includes terms from Luganda, Runyankole, Acholi, Ateso, Lugbara, Runyoro
Designed to test code-switching and local linguistic awareness

Dataset Structure

Each instance is a structured QA pair:

{
  "id": "22410",
  "category": "Slang & Local Expressions",
  "question": "What does the term 'rolex' mean in Ugandan slang?",
  "answer": "A 'rolex' is a popular street food in Uganda..."
}

Fields:

id: Unique identifier
category: Cultural domain
question: Textual query for evaluation
answer: Ground-truth culturally contextual response

Coverage & Categories

Category	Category
Education	Attires & Dress Culture
Ugandan Herbs	Value Addition
Media	Customs
Economy	Slang & Expressions
Notable Key Figures	Festivals
Literature	Food Culture
Architecture	Geography
Folklore	Demographics
Language	History
Religion	Traditions & Rituals
Social Norms	Streetlife
Music	Sports

Total Examples: 1,039
Average per Category: ~43
Split: test only (evaluation use)

Curation Goals

The UCCB Suite was curated to go beyond surface-level recall. Goals include:

Testing deeper reasoning over Uganda-specific content
Assessing awareness of social dynamics and local idioms
Ensuring inclusivity across ethnic, gender, and regional dimensions
Creating a trusted QA evaluation benchmark for Ugandan use cases in health, education, governance, and AI fairness

Source Data

Expert-authored and fact-checked using:
- Ugandan academic literature and historical texts
- National archives (e.g., UBOS, UCC)
- Cultural institutions (e.g., Uganda Museum, Cross Cultural Foundation)
- National media (e.g., Observer Media)
- Peer-reviewed ethnographic and linguistic studies

Annotation & Review Process

Multi-stage human-in-the-loop review
Flagged and reviewed with a rubric assessing:
- ✅ Factual Accuracy
- ✅ Clarity and Cultural Specificity
- ✅ Representation and Diversity
- ✅ Bias Detection and Mitigation

Known Limitations

Representational Bias: Slight overrepresentation of central and western cultures
Urban-Centric Influence: Kampala-related examples may be more frequent
Cultural Dynamism: Slang and media categories may age quickly
Scope Limitation: While wide-reaching, it cannot exhaustively cover all ~50 ethnic groups

Licensing

This dataset is licensed under:

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) View License

Citation

@misc{uccb_2025,
  author    = {Lwanga Caleb and Gimei Alex and Kavuma Lameck and Kato Steven Mubiru and Roland Ganafa and Sibomana Glorry and Atuhaire Collins and JohnRoy Nangeso and Bronson Bakunga},
  title     = {The Ugandan Cultural Context Benchmark (UCCB) Suite},
  year      = {2025},
  url       = {https://huggingface.co/datasets/CraneAILabs/UCCB}
}

Contributions

Thanks to the core team at AI Studio Uganda.

Usage

from datasets import load_dataset

# Load the extended UCCB dataset
dataset = load_dataset("aistudioug/UCCB-Suite")

# View a few examples
sample = dataset["test"].select(range(3))
for row in sample:
    print("Category:", row["category"])
    print("Question:", row["question"])
    print("Answer:", row["answer"])
    print("-" * 30)