Datasets:

blue-machines
/

tokenwise_lid_all_scripts_dataset

license: other
task_categories:
  - token-classification
language:
  - en
  - hi
  - bn
  - gu
  - kn
  - ml
  - mr
  - or
  - ta
  - te
tags:
  - language-identification
  - token-classification
  - indic
  - code-mixed
  - romanized
pretty_name: Tokenwise LID All Scripts Dataset
size_categories:
  - 1M<n<10M
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: text
      dtype: string
    - name: token_types
      dtype: string
    - name: language
      dtype: string
    - name: language_switch
      dtype: string
    - name: tokens
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 557958553
      num_examples: 1965547
  download_size: 176434491
  dataset_size: 557958553

blue-machines/tokenwise_lid_all_scripts_dataset

Word/token-level LID training corpus (token_lid_combined_dataset.csv).

Token labels are pipe-separated (|) in token_types / tokens:

LID_WORD
IGNORE_ENTITY

Columns

text
token_types
language
language_switch
tokens
source

Stats

Rows: 1,965,547
Languages: Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Oria, Tamil, Telugu

Load

from datasets import load_dataset
ds = load_dataset("blue-machines/tokenwise_lid_all_scripts_dataset")
print(ds["train"][0])