README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: string
    - name: ja
      dtype: string
    - name: en
      dtype: string
  splits:
    - name: train
      num_bytes: 17758809
      num_examples: 147876
  download_size: 10012915
  dataset_size: 17758809
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - translation
language:
  - ja
  - en
pretty_name: tanaka-corpus
size_categories:
  - 100K<n<1M

HF Datasets version of Tanaka Corpus.

Preprocess for HF Datasets

以下の内容でオリジナルデータを前処理しました。

wget ftp://ftp.edrdg.org/pub/Nihongo/examples.utf.gz
gunzip examples.utf.gz

import re
from pathlib import Path

from more_itertools import chunked

import datasets as ds

data = []
with Path("examples.utf").open() as f:
    for row, _ in chunked(f, 2):
        ja, en, idx = re.findall(r"A: (.*?)\t(.*?)#ID=(.*$)", row)[0]
        data.append(
            {
                "id": idx,
                "ja": ja.strip(),
                "en": en.strip(),
            }
        )

dataset = ds.Dataset.from_list(data)
dataset.push_to_hub("hpprc/tanaka-corpus")