cis-lmu/GlotStoryBook-MT

Dataset Description

Machine Translation (MT) version of Story Books for 180 ISO-639-3 codes (190 variety of languages). Original dataset: cis-lmu/GlotStoryBook.

This dataset consisted of 4 publishers:

1. asp: African Storybook
2. pb: Pratham Books
3. lcb: Little Cree Books
4. lida: LIDA Stories

• GitHub Repository: github
• Paper: paper
• Point of Contact: amir@cis.lmu.de

Usage (HF Loader)

from datasets import load_dataset 
dataset = load_dataset('cis-lmu/GlotStoryBook-MT', 'en')
print(dataset['test'][0]) # First row data for en versus other languages

Download

If you are not a fan of the HF dataloader, download it directly:

First, check out the directory of language of your interest (for example, 'en'):

https://huggingface.co/datasets/cis-lmu/GlotStoryBook-MT/tree/main/global/en

Then, download the pair of your interest (en-fa here):

! wget https://huggingface.co/datasets/cis-lmu/GlotStoryBook-MT/blob/main/global/en/en-fa.csv

You can also clone the whole directory:

! git clone https://huggingface.co/datasets/cis-lmu/GlotStoryBook-MT

Format

Each sentence is included in a list because for some texts in the source and target languages, two versions of translations exist. However, these lists are converted to strings in this dataset.

You can bring them back to be lists again.

For example:

from datasets import load_dataset
from ast import literal_eval

data_en = load_dataset("cis-lmu/GlotStoryBook-MT", 'en')

# convert the datasets object to pandas (optional)
df_en = data_en['test'].to_pandas()

# you can also use eval function for each entry.
df_en['source_sentences'] = df_en['source_sentences'].apply(literal_eval)
df_en['target_sentences'] = df_en['target_sentences'].apply(literal_eval)
df_en['source_files'] = df_en['source_files'].apply(literal_eval)
df_en['target_files'] = df_en['target_files'].apply(literal_eval)
df_en.head()

License and Copyright

We do not own any of the text from which these data has been extracted. All the files are collected from the repository located at https://github.com/global-asp/. The source repository for each text and file is stored in the original dataset: cis-lmu/GlotStoryBook. Each file in the dataset is associated with one license from the CC family. The licenses include 'CC BY', 'CC BY-NC', 'CC BY-NC-SA', 'CC-BY', 'CC-BY-NC', and 'Public Domain'. We also license the code, actual packaging and the metadata of these data under the cc0-1.0.

Citation

If you use any part of this code and data in your research, please cite it (along with https://github.com/global-asp/) using the following BibTeX entry. This work is part of the GlotLID project.

@inproceedings{
  kargaran2023glotlid,
  title={{GlotLID}: Language Identification for Low-Resource Languages},
  author={Kargaran, Amir Hossein and Imani, Ayyoob and Yvon, Fran{\c{c}}ois and Sch{\"u}tze, Hinrich},
  booktitle={The 2023 Conference on Empirical Methods in Natural Language Processing},
  year={2023},
  url={https://openreview.net/forum?id=dl4e3EBz5j}
}