---
annotations_creators:
- synthetic
language_creators:
- machine-generated
language:
- ak
- twi
license: cc-by-4.0
multilinguality:
- monolingual
pretty_name: Akan POS Tagging Dataset - 10 Million Sentences
size_categories:
- 1M<n<10M
source_datasets:
- original
task_categories:
- text-classification
- token-classification
task_ids:
- part-of-speech
---
# Akan POS Tagging Dataset - 10 Million Sentences

## Dataset Description

The dataset includes sentences and part-of-speech tags and is aimed at supporting the development of POS tagging models for the Akan language.
This work demonstrates that data limitations for low-resource languages can be overcome using artificial data generation techniques.

## Dataset Structure

### Data Fields

- `sentence`: The Akan sentence text
- `pos_sequence`: Part-of-speech tags for the sentence. The [POS-tags.pdf](./POS-tags.pdf) includes full meanings and descriptions of the POS tags from the dataset.
- `word_count`: Number of words in the sentence

### Data Splits

The dataset is provided as a single training split containing 10 million sentence-POS tag pairs.

### Dataset Statistics

- **Total sentences**: 10,000,000
- **Language**: Akan (Akan)
- **Task**: Part-of-Speech Tagging
- **Data Generation**: Artificial data generation techniques

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("michsethowusu/aka-pos-10m")

# Access the data
train_data = dataset['train']
print(f"Number of examples: {len(train_data)}")

# Example usage
example = train_data[0]
print(f"Sentence: {example['sentence']}")
print(f"POS Tags: {example['pos_sequence']}")
print(f"Word Count: {example['word_count']}")
```

## Research Applications

This dataset is particularly valuable for:

- Developing POS tagging models for Akan/Akan language
- Research in low-resource language processing
- Studies on artificial data generation for NLP
- Cross-lingual transfer learning experiments


## Acknowledgement

The seed dataset was obtained from the following source:

```bibtex
@misc{Beermann2018,
  author       = {Dorothee Beermann},
  title        = {The TypeCraft Akan corpus, Release 1.0},
  year         = {2018},
  howpublished = {\url{https://typecraft.org/tc2wiki/The_TypeCraft_Akan_Corpus}},
  note         = {TypeCraft – The Interlinear Text Repository}
}
```

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{aka_pos_10m,
  title={Akan POS Tagging Dataset - 10 Million Sentences},
  author={Mich-Seth Owusu},
  year={2025},
  publisher={Hugging Face},
  url={michsethowusu/twi-fante-sentences-parts-of-speech-pos_tagged-10m}
}
```

## Collaboration

We invite researchers working in low-resource language processing to collaborate on research papers using this dataset.
If you're interested in co-authoring a paper, please reach out!

## License

[Specify your license here - e.g., Apache 2.0, MIT, CC BY 4.0, etc.]

## Acknowledgments

This work demonstrates the potential of artificial data generation techniques for overcoming data limitations
in low-resource language processing, specifically for African languages.