ReliableAI/Irish-English-Parallel-Collection

UCCIX's English-Irish Parallel Textual Corpus

Dataset Summary

This parallel English-Irish text dataset includes data from various sources such as paracrawl.eu, ECLR.

This dataset is feed to the English-centric pre-trained LLM at the start of continual pre-training, with the hypothesis to allow the LLM to draw the connections between the two languages easier, before learning on mono Irish data.

Dataset Sources

Source	Description	Statistics	Note
paracrawl.eu	Translation sentences crawl from the Internet	A subset of 50,095 Translation Units.	Denoted as gaparacrawl:{id}.
The Coimisineir Teanga Bilingual Web Corpus (Processed) – ELRC-SHARE	Web content from the Language Commissioner's Office.	321 Translation Units.	Denoted as eclr:{id}.
The Coimisineir Teanga Bilingual Corpus of Reference Documents (Processed) – ELRA Catalogue	General Reference content from the Language Commissioner's Office.	1,337 Translation Units.	Denoted as eclr:{id}.
The Coimisineir Teanga Bilingual Corpus of Reports and Press Releases (Processed) – ELRC-SHARE	Reports and Press Release data from the Language Commissioner's Office.	2,323 Translation Units.	Denoted as eclr:{id}.
Bilingual Web-Corpus (Processed) – ELRC-SHARE	A web corpus crawled from http://www.citizensinformation.ie. Contains 10,297 parallel sentences of English/Irish that have undergone manual cleaning. May be reproduced and/or re-used free of charge subject to the latest PSI licence, Creative Commons Attribution 4.0 International (CC BY 4.0).	10,297 Translation Units.	Denoted as eclr:{id}.
English-Irish website parallel corpus (Processed) – ELRC-SHARE	This is a parallel corpus of bilingual texts crawled from multilingual websites, which contains 1134 TUs. Manual validation has been performed on a sample of the data.	1,134 Translation Units.	Denoted as eclr:{id}.

Load the dataset

You can load the dataset in Python using the datasets library:

from datasets import load_dataset
dataset = load_dataset("ReliableAI/Irish-English-Parallel-Collection")

Considerations for Using the Data

As this corpus is a collection of multiple datasets from Internet sources, it might still contain personal and sensitive information. Before using this dataset for any purpose, including training deep learning models, this should be to be taken into account.

Dataset Structure

{
    "text": ...,
    "eng_text": ...,
    "id": ...,
}

Citation

To cite this dataset, please use:

@misc{tran2024uccix,
      title={UCCIX: Irish-eXcellence Large Language Model}, 
      author={Khanh-Tung Tran and Barry O'Sullivan and Hoang D. Nguyen},
      year={2024},
      eprint={2405.13010},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}