Yue-Wiki-PL-BERT Dataset

Overview

This dataset contains processed text data from Cantonese Wikipedia articles, specifically formatted for training or fine-tuning BERT-like models for Cantonese language processing. The dataset is created by hon9kon9ize and contains approximately 176,177 rows of training data.

Description

The Yue-Wiki-PL-BERT dataset is a structured collection of Cantonese text data extracted from Wikipedia, with each entry containing:

• id: A unique identifier for each entry
• url: The source Wikipedia URL (zh-yue.wikipedia.org)
• title: The title of the Wikipedia article in Cantonese
• input_ids: Tokenized numerical representations of the text content
• phonemes: Phonetic representations of the text using Cantonese romanization of Jyutping

The dataset is primarily designed for pre-training or fine-tuning language models on Cantonese text, with a focus on both character-level and phonetic-level understanding of the language.

Background & Purpose

Cantonese is a language spoken by approximately 85 million people worldwide, primarily in Hong Kong, Macau, Guangdong, and overseas Cantonese communities. Despite its significant speaker population, Cantonese is underrepresented in NLP research compared to Mandarin.

This dataset aims to address this gap by providing a structured corpus for developing Cantonese-specific language models. The inclusion of phoneme data suggests a focus on models that can understand both the written characters and pronunciation patterns of Cantonese, which is particularly valuable given Cantonese's complex tonal system and the frequent use of colloquial characters that aren't found in standard Chinese.

Dataset Statistics

• Size: 176,177 rows
• File Size: 93.5 MB (Parquet converted)
• Format: Parquet
• Text Modality: Cantonese text with phonetic transcriptions

Usage

This dataset can be used for:

1. Pre-training Cantonese language models
2. Fine-tuning existing multilingual models for Cantonese-specific tasks
3. Research on phoneme-aware language models
4. Developing Cantonese NLP applications

Loading the Dataset

Using the Datasets library:

from datasets import load_dataset

# Load from Hugging Face Hub (if available)
dataset = load_dataset("hon9kon9ize/yue-wiki-pl-bert")

# Or load locally if downloaded
dataset = load_dataset("parquet", data_files="path/to/dataset.parquet")

Using pandas:

import pandas as pd

df = pd.read_parquet("path/to/dataset.parquet")

Data Format

Each row in the dataset contains:

• id: String - Unique identifier
• url: String - URL pointing to the source Wikipedia article
• title: String - Title of the article in Cantonese characters
• input_ids: Sequence - Numerical token IDs for the text content
• phonemes: Sequence - Romanized phonetic representations of the text

Example entries include articles with titles:

• 香港 (Hong Kong)
• 愛因斯坦 (Einstein)
• 粵語 (Cantonese )
• 中文 (Chinese language)
• 太平山 (Victoria Peak)
• 北京 (Beijing)

License

Since this dataset is derived from Wikipedia content, it is likely subject to the same license as Wikipedia content - Creative Commons Attribution-ShareAlike 3.0 Unported License (CC BY-SA 3.0) or the GNU Free Documentation License (GFDL).

Users should verify the specific license information with the dataset creator.

Citation

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

@dataset{hon9kon9ize2024,
  author       = {hon9kon9ize},
  title        = {Yue-Wiki-PL-BERT: A Cantonese Wikipedia Dataset for Phoneme-aware Language Modeling},
  year         = {2024},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/hon9kon9ize/yue-wiki-pl-bert}}
}

Contribution

To contribute to this dataset:

1. Fork the repository on Hugging Face Hub
2. Make your changes or additions
3. Submit a pull request with detailed description of your changes
4. Contact the original creator (hon9kon9ize) for permissions and guidance

Limitations

• The dataset is focused exclusively on Wikipedia content, which may not represent all varieties of written Cantonese
• Wikipedia content tends to be more formal than everyday Cantonese usage
• The phonetic representations may follow a specific romanization system that should be documented

Acknowledgements

Thanks to the Cantonese Wikipedia contributors who created the original content that this dataset is based on.

Contact

For more information, please contact the dataset creator hon9kon9ize through HuggingFace.