README.md

metadata

dataset_info:
  features:
    - name: src_lang
      dtype:
        class_label:
          names:
            '0': eng_Latn
            '1': ben_Beng
            '2': guj_Gujr
            '3': hin_Deva
            '4': kan_Knda
            '5': mal_Mlym
            '6': mar_Deva
            '7': tam_Taml
            '8': tel_Telu
    - name: tgt_lang
      dtype:
        class_label:
          names:
            '0': eng_Latn
            '1': ben_Beng
            '2': guj_Gujr
            '3': hin_Deva
            '4': kan_Knda
            '5': mal_Mlym
            '6': mar_Deva
            '7': tam_Taml
            '8': tel_Telu
    - name: src_text
      dtype: string
    - name: tgt_text
      dtype: string
    - name: score
      dtype: float64
  splits:
    - name: train
      num_bytes: 1698080949
      num_examples: 3990964
  download_size: 817154884
  dataset_size: 1698080949
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
task_categories:
  - translation
language:
  - bn
  - hi
  - ta
  - te
  - mr
  - ml
  - kn
  - gu
size_categories:
  - 1M<n<10M

A curated subset of Bharat Parallel Corpus Collection (BPCC) for 8 Indian languages. Translation pairs are filtered with LABSE score(>0.9) and further preprocessed. Useful for training high-quality translation models.