TIgrinyaLargeTextDataset

Dataset Description

TIgrinyaLargeTextDataset is a large-scale collection of Tigrinya language articles designed for training Large Language Models (LLMs) and other Natural Language Processing (NLP) tasks. This dataset addresses the critical need for high-quality Tigrinya language resources in the machine learning community.

Dataset Summary

• Language: Tigrinya
• Task Categories: Language Modeling, Text Generation, NLP Research
• Size: 12,374 articles
• Total Words: 5,935,239
• Total Characters: 28,855,522
• Vocabulary Size: 553,989 unique words
• Unique Texts: 12,299
• Time Period: 2020-2024

Dataset Structure

Data Format

The dataset is provided in JSONL (JSON Lines) format, where each line contains a single article entry.

Data Fields

Each article entry contains the following fields:

• title: The title of the article
• content: The content of the article
• category: The category/topic classification (currently unclassified)
• source: The source website or publication

Data Statistics

• Average characters per text: 2,331.9
• Average words per text: 479.7
• Average sentences per text: 32.1
• Character distribution: 76.6% Tigrinya characters, 0.8% digits, 1.1% punctuation

Data Sources

The articles are collected from various Tigrinya language sources, including:

• Haddas Eritra newspaper
• Other Tigrinya websites

Supported Tasks

This dataset can be used for various NLP tasks including:

• Language Modeling: Training autoregressive language models
• Text Generation: Fine-tuning models for Tigrinya text generation
• Machine Translation: As source or target data for translation models
• Text Classification: Once categories are properly labeled
• Named Entity Recognition: Training NER models for Tigrinya
• Sentiment Analysis: Developing sentiment analysis tools

Languages

• Tigrinya: Primary language of the dataset

Dataset Creation

Curation Rationale

This dataset was created to address the scarcity of large-scale, high-quality Tigrinya language resources for machine learning applications. Tigrinya is an under-resourced language, and this collection aims to support research and development in Tigrinya NLP.

Source Data

Data Collection

Articles were collected from publicly available Tigrinya language websites and publications, spanning the years 2020-2024.

Data Processing

The text has been processed and structured into a consistent JSONL format for easy integration into machine learning pipelines.

Considerations for Using the Data

Social Impact of Dataset

This dataset contributes to language preservation and technological inclusion by providing resources for Tigrinya language processing. It enables the development of AI tools that can serve Tigrinya-speaking communities.

Discussion of Biases

Users should be aware that the dataset reflects the perspectives and biases present in the source materials. The articles come from specific news sources and websites, which may have particular editorial viewpoints.

Other Known Limitations

• Categories are not yet classified, limiting supervised learning applications
• The dataset covers a specific time period (2020-2024)
• Source diversity may be limited to available online Tigrinya content

Additional Information

Dataset Curators

Mewael Tsegay Desta, Finetuning LLMs on low-resource languages, the case of Tigrinya.

Licensing Information

This dataset is released under the MIT License.

Citation Information

@dataset{tigrinya_large_text_dataset_2024,
  title={TIgrinyaLargeTextDataset},
  author={Mewael_Tsegay_Desta},
  year={2024},
  url={https://huggingface.co/datasets/mewaeltsegay/TigrinyaLargeText}
}

Contributions

Thanks to the Tigrinya language community and the publishers who made their content available for research purposes.