Updated version
Browse files# Niger Delta Climate-Health Dialogues: Hausa, Yoruba & Igbo
**A parallel conversational dataset for climate adaptation and public health in under-resourced Nigerian languages.**
### Why I built this
During my **Metabolism of Carbohydrates** exam, I observed students from different ethnic groups naturally switching between Igbo, Hausa, Yoruba, and Pidgin while asking for answer booklets. That moment made me realise how language shapes understanding — especially for critical topics like health in the face of climate change.
People in the Niger Delta face real, everyday challenges: flooding-induced malaria, antenatal care barriers during floods, respiratory problems from gas flaring soot, hypertension from economic stress caused by environmental damage, child nutrition issues after dirty flood water, mental health impacts from climate-related hardship, and difficulty accessing care during disasters. Yet high-quality conversational health data in local languages remains extremely scarce.
I created this dataset to help bridge that gap — so people can receive practical health information in the languages they understand best.
### What this dataset is trying to do
This dataset provides ** natural doctor-patient dialogues** in Hausa, Yoruba, and Igbo that reflect real climate-health adaptation scenarios in the Niger Delta. The goal is to support the development of inclusive AI systems that can communicate effectively with Nigerians in their own languages during climate-related health challenges.
### Conversative examples
- A mother worried about her child getting malaria after flood water
- A pregnant woman struggling with antenatal access due to flooding and distance
- A patient experiencing respiratory issues from gas flaring soot
- Someone dealing with hypertension linked to economic stress from environmental damage
### Negative space
This dataset deliberately avoids generic or unrelated health topics. It focuses only on the intersection of climate change and health in the Niger Delta.
### Edge cases
Some dialogues include symptom evolution, patient hesitation, and real-world barriers (cost, distance, trust in medicine) to make the conversations more realistic and useful for adaptive AI.
### Code-mixed language
The dataset is **clean parallel** (no code-mixing or Pidgin) to ensure high quality and ease of use for multilingual model training. Each conversation is provided in full parallel form across Hausa, Yoruba, Igbo, and English.
### Languages
- Hausa
- Yoruba
- Igbo
- English (reference)
All samples are parallel across these languages.
### What changed during development
- Started with Pidgin but removed it to focus on the three major national languages for better quality and wider usability.
- Shifted focus from general health to **climate-health adaptation** to better address an underserved domain.
- Expanded from 20 to 40 rows after feedback to increase diversity and depth.
### Dataset structure
- `id`
- `topic`
- `region`
- `english_question`
- `english_answer`
- `hausa_question`
- `hausa_answer`
- `yoruba_question`
- `yoruba_answer`
- `igbo_question`
- `igbo_answer`
### Key observation
Conversations in local languages feel more natural and empathetic. Patients express concerns more freely, and doctors can give clearer, culturally relevant advice when using Hausa, Yoruba, or Igbo.
### Limitations
- Synthetic dialogues (no real patient data)
- Focused on Rivers State / Niger Delta — not representative of all of Nigeria
- Limited to 40 rows (can be expanded)
### Ethical note
All dialogues are fictional but grounded in real public health issues. The dataset aims to improve access to health information and does not contain any personally identifiable information.
**Created for the Uncharted Data Challenge by Adaption Labs**
**License**: CC-BY-4.0
**Creator**: Edidiong (Rivers State, Nigeria) — April 2026