hsilvosa/hsc-wuxia-full

HSC Wuxia Full Dataset

This dataset is the complete master parallel corpus of Chinese-English sentence pairs developed for the Bachelor's Thesis (Trabajo de Fin de Grado - TFG) titled:

"Enfoques de traducción automática con modelos de lenguaje en obras wuxia" (Degree in Data Science and Engineering, Universidade da Coruña)

It contains the full set of clean, aligned parallel sentence pairs extracted from the structural and semantic alignment of Wuxia novels.

Corpus Sources

The texts are extracted from three popular web novels of the Wuxia / Xianxia genre:

• AWE: A Will Eternal (一念永恒)
• CONDOR: The Condor Trilogy (射雕三部曲 - Jin Yong - 3 volumes)
• GU: Reverend Insanity (蛊真人 - Guzhenren collection)

Dataset Splits & Exact Row Counts

The dataset is structured as follows:

Split	Description	Exact Number of Pairs
train	Full training set	417,208
validation	Standard validation split	52,151
test	Evaluation and test split	52,151
Total		521,510

Corpus Construction Pipeline

1. Segmentation: Tokenizes Chinese and English chapters into structural paragraphs using regular expressions custom-tailored for historical Chinese chapter tags (e.g., 第...章, 第...回).
2. Vector Space Encoding: Maps segments into a shared multilingual dense vector space using LaBSE (Language-Agnostic BERT Sentence Embeddings).
3. Sentence Alignment: Employs a dynamic programming alignment matrix filtered by cosine similarity thresholds to pair high-confidence parallel sentences.

Dataset Format

The dataset is structured as a Hugging Face DatasetDict ready for direct integration:

• zh: Original sentence in Chinese.
• en: Translated sentence in English.

from datasets import load_dataset
dataset = load_dataset("HSilvosa/hsc-wuxia-full")

Project Repository

This dataset was developed in the following project repository:

• GitHub Repository: HugoSilvosa/hsc-mtl-wuxia