metadata

license: cc-by-nc-4.0
configs:
  - config_name: default
    data_files:
      - split: bam
        path: data/bam-*
      - split: boz
        path: data/boz-*
      - split: bom
        path: data/bom-*
      - split: myk
        path: data/myk-*
      - split: sgh
        path: data/sgh-*
      - split: snk
        path: data/snk-*
dataset_info:
  features:
    - name: original
      dtype: string
    - name: cleaned
      dtype: string
    - name: space_insertion
      dtype: string
    - name: char_delete
      dtype: string
    - name: char_swap
      dtype: string
    - name: word_swap
      dtype: string
    - name: word_delete
      dtype: string
    - name: word_blank
      dtype: string
  splits:
    - name: bam
      num_bytes: 51067075
      num_examples: 100000
    - name: boz
      num_bytes: 352249
      num_examples: 254
    - name: bom
      num_bytes: 3097936
      num_examples: 3085
    - name: myk
      num_bytes: 22544040
      num_examples: 32916
    - name: sgh
      num_bytes: 39044992
      num_examples: 56126
    - name: snk
      num_bytes: 2510973
      num_examples: 8181
  download_size: 85681755
  dataset_size: 118617265
task_categories:
  - text-classification
  - question-answering
  - zero-shot-classification
language:
  - bm
  - ses
  - snk
  - bmq
  - bze
  - myk
pretty_name: mlnoisy
size_categories:
  - 100K<n<1M
version: 0.1.1

ml.Datx - Noisy

This dataset contains texts in six Malian languages with various noise functions applied. It is designed for use with Hugging Face's NLP tools and models.

Languages Included

Bambara
Bomu
Bozo
Mamama
Songhay
Soninke

Noise Functions Applied

The following noise functions have been applied to the texts:

Character Swap: Randomly swaps adjacent characters.
Random Space Insertion: Inserts random characters at random positions.
Random Character Deletion: Deletes random characters.
Random Word Position Swapping: Substitutes random characters with other random characters.
Word Masking: Random word masking using BLANK keyword

Dataset Structure

The dataset is structured as follows:

data/lang_id_output.json`: Contains a list based dictionary representation of the data

Each data instance contains the followings original, space_insertion, char_delete, char_swap, word_swap, word_delete, word_blank, cleaned, for example:

[
    {
        "original": "Yesu Kirisita Tuura Wema TUURA WEMA",
        "space_insertion": " Ye su Kiris i ta Tu u r a  Wem a  TUU RA   WEM A",
        "char_delete": "Ysu isia Tuurema TURAWMA",
        "char_swap": "eYsu iKisriat Turua Wma eTUURA WMEA",
        "word_swap": "Yesu Kirisita Tuura Wema WEMA TUURA",
        "word_delete": "Yesu Kirisita Tuura Wema TUURA WEMA",
        "word_blank": "Yesu Kirisita Tuura Wema BLANK WEMA",
        "cleaned": "Yesu Kirisita Tuura Wema TUURA WEMA"
    }
...
]

Usage

To use this dataset with Hugging Face, you can load it using the datasets library:

from datasets import load_dataset

dataset = load_dataset('mlsftwrs/mlnoisy')

Sources

Citation

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

@misc{diarra2025mldatxnoisy,
    title={ml.Datx:Noisy -  Malian Noisy Dataset for LLM Training},
    author={
        Sebastien Diarra and
        Nene Tangara and
        Seydou Diallo
    },
    url={https://mlsftwrs.ml/project/mldatx},
    year={2025}
}

Contact

For any questions or issues, please contact:

[email protected].