bangboom/chinese-translation-data

Chinese Translation Data (WebTrans)

This dataset contains training data for Chinese text translation tasks, including classical-to-modern Chinese, modern-to-classical Chinese, punctuation addition, and poetry translation. The data is in binarized fairseq format ready for training sequence-to-sequence models.

Dataset Structure

├── bpe_file/              # BPE model files
│   ├── addpunc.bpe        # BPE codes for punctuation task
│   ├── new.bpe            # BPE codes for modern Chinese
│   ├── old.bpe            # BPE codes for classical Chinese
│   └── shici.bpe          # BPE codes for poetry
├── dataflow_addpunc/      # Binarized data for punctuation addition
│   ├── dict.src.txt       # Source vocabulary
│   ├── dict.tgt.txt       # Target vocabulary
│   └── {train,valid,test}.src-tgt.{src,tgt}.{bin,idx}
├── dataflow_new2old/      # Binarized data: modern → classical Chinese
│   ├── dict.new.txt / dict.old.txt
│   └── {train,valid,test}.new-old.{new,old}.{bin,idx}
├── dataflow_old2new/      # Binarized data: classical → modern Chinese
│   ├── dict.new.txt / dict.old.txt
│   ├── tag
│   └── {train,valid,test}.old-new.{new,old}.{bin,idx}
├── dataflow_shici/        # Binarized data: poetry translation
│   ├── dict.key.txt / dict.sen.txt
│   └── {train,valid,test}.key-sen.{key,sen}.{bin,idx}
└── dictionary/            # Chinese dictionaries
    ├── dict.txt
    ├── 中国历史地名词典.txt   # Chinese historical place names
    ├── 古代人名（25w）.txt    # Ancient person names (250K)
    └── 成语（5W）.txt        # Chinese idioms (50K)

Tasks

Directory	Task	Source → Target
dataflow_new2old	Modern to Classical Chinese	new → old
dataflow_old2new	Classical to Modern Chinese	old → new
dataflow_addpunc	Punctuation Addition	src → tgt
dataflow_shici	Poetry Translation	key → sen

Usage

This data is in fairseq binarized format. To use with fairseq:

fairseq-train dataflow_old2new \
  --save-dir checkpoints \
  --arch transformer_iwslt_de_en \
  --optimizer adam \
  --lr 0.0005 \
  --max-tokens 4096

License

Apache-2.0