---
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](https://bpcc.ai4bharat.org/) (BPCC) for 8 Indian languages. 
Translation pairs are filtered with LABSE score(>0.9) and further preprocessed.
Useful for training high-quality translation models.