Datasets:

agentlans
/

en-zhtw

language:
  - en
  - zh
multilinguality:
  - translation
size_categories:
  - 100K<instances<1M
source_datasets:
  - jslin09/news_commentary_tw
  - zetavg/coct-en-zh-tw-translations-twp-300k
task_categories:
  - translation
pretty_name: English ↔ Traditional Chinese Translation Dataset
dataset_info:
  features:
    - name: en
      dtype: string
    - name: zhtw
      dtype: string
    - name: translation_quality
      dtype: float32
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: null
      num_examples: 333243
    - name: test
      num_bytes: null
      num_examples: 83311
  download_size: null
  dataset_size: null

English ↔ Traditional Chinese Translation Dataset

This dataset is a curated combination of two high-quality parallel corpora for English to Traditional Chinese (zh-TW) translation. It is designed for training, fine-tuning, or evaluating machine translation models, especially in academic or production settings.

📚 Dataset Overview

Source datasets:
- jslin09/news_commentary_tw
- zetavg/coct-en-zh-tw-translations-twp-300k
Filtering criteria:
- Sentence pairs scored using LaBSE (Language-Agnostic BERT Sentence Embedding) to assess translation quality.
- Only pairs with a LaBSE cosine similarity > 0.7 were retained.
- English sentence lengths between 5 and 500 characters were kept.
- Exact duplicates across datasets were removed.

📊 Dataset Statistics

Source Dataset	Train Samples	Test Samples	Total
`jslin09/news_commentary_tw`	186 232	46 558	232 790
`zetavg/coct-en-zh-tw-translations-twp-300k`	147 011	36 753	183 764
Combined Total	333 243	83 311	416 554

📝 Example

{
  "en": "Alzheimer's disease or Parkinson's disease, whose causes are unclear, are both connected to the decline of the brain, notes Hong-hsin Lee.",
  "zhtw": "像至今原因不明的老年癡呆症、帕金森氏症等，都是腦部退化引起的」，李宏信指出。",
  "translation_quality": 0.7144,
  "source": "zetavg/coct-en-zh-tw-translations-twp-300k"
}

🛠️ Intended Use

This dataset is intended for:

Training and evaluating English ↔ Traditional Chinese neural machine translation (NMT) models.
Sentence similarity and alignment benchmarking.
Linguistic and domain-specific translation research.

📌 Notes

Translation quality scores are provided to allow for additional filtering based on confidence.
The dataset can be further split or rebalanced depending on downstream application needs.