---
dataset_info:
  features:
  - name: source_type
    dtype: large_string
  - name: source
    dtype: large_string
  - name: text
    dtype: large_string
  - name: generated_question
    dtype: large_string
  splits:
  - name: train
    num_bytes: 401345393
    num_examples: 226300
  download_size: 213137497
  dataset_size: 401345393
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
language:
- en
license: cc-by-nc-4.0
tags:
- ghana
- chat
- nlp
datasets:
- ghananlpcommunity/ghana-chat
---

# Ghana Chat Corpus — English

English-only base dataset extracted from
[ghananlpcommunity/ghana-chat](https://huggingface.co/datasets/ghananlpcommunity/ghana-chat).

Each conversation is split into the first user message (`generated_question`) and all
assistant responses joined by `\n\n` (`text`).

## Schema

| Column | Description |
|--------|-------------|
| `source_type` | Source category (e.g., news, parliament) |
| `source` | Original source name |
| `text` | English conversation (all assistant responses joined) |
| `generated_question` | First user message in the conversation |

## Intended Use

Base English corpus for translation pipelines and multilingual LLM training.

## Citation

```
@software{ghana-chat-corpus,
  title = {Ghana Chat Corpus — English},
  author = {Owusu, Mich-Seth and Ghana NLP Community},
  year = {2026},
}
```