AfricaBias-SW-FR: African Bias Detection Dataset for Swahili and French

Dataset Summary

AfricaBias-SW-FR is a multilingual bias detection dataset containing 35,285 annotated sentences in Swahili (24,289) and French (10,996). It was developed by Makini AI Limited to address the critical shortage of African-language bias benchmarks. Each sentence is labeled across four bias categories: neutral, stereotype, counter-stereotype, and derogation, spanning eight social domains including household and care, livelihoods and work, governance, health, and education.

This is the first publicly released large-scale gender and social bias detection dataset for Swahili. It was produced as a Gates Foundation-supported deliverable through the AfriLabs AI Accelerator program.

Languages

• Swahili (sw) — 24,289 instances
• French (fr) — 10,996 instances

Label Classes

Label	Count	Percentage
Neutral	24,220	68.6%
Stereotype	9,605	27.2%
Counter-Stereotype	1,288	3.7%
Derogation	172	0.5%

Dataset Fields

Field	Description
id	Unique instance ID
language	Language code: sw or fr
text	The sentence to be classified
bias_label	Primary label: neutral, stereotype, counter-stereotype, derogation
target_gender	Gender target: male, female, mixed, neutral, unknown
domain	Thematic domain
theme	Fine-grained theme
source_type	Source type: news, social_media, annotated, synthetic, etc.
source_ref	URL or reference to original source
safety_flag	Content safety: safe, sensitive, reject
pii_removed	Whether PII removal was applied

Usage

from datasets import load_dataset

dataset = load_dataset("Daudipdg/africabias-sw-fr")

License

CC BY-NC 4.0 — Free for research and non-commercial use. Commercial use requires explicit permission from Makini AI Limited.

Citation

@dataset{wachira2026africabias,
  title     = {AfricaBias-SW-FR: African Bias Detection Dataset for Swahili and French},
  author    = {Wachira, David Maina},
  year      = {2026},
  version   = {1.0},
  publisher = {Algedi Inteligence Labs},
  url       = {https://huggingface.co/datasets/Daudipdg/africabias-sw-fr},
  
}

Contact

David Maina Wachira — dave@makini.tech — https://makini.tech