metadata
license: apache-2.0
dataset_info:
features:
- name: inputs
dtype: string
- name: targets
dtype: string
- name: language
dtype: string
- name: language_code
dtype: string
- name: annotation_type
dtype: string
- name: user_id
dtype: string
splits:
- name: train
num_bytes: 11918662
num_examples: 6633
- name: test
num_bytes: 308222
num_examples: 250
download_size: 4492541
dataset_size: 12226884
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
task_categories:
- question-answering
language:
- bn
pretty_name: (Subset of) Aya dataset translated to Bengali
size_categories:
- 1K<n<10K
aya_dataset_ben_translated is a subset of the aya_dataset, with some modifications. In particular, the original data points in Bengali (indicated by the language or language_code columns) are retained. In addition, the English and Hindi data points are translated into Bengali using Google Cloud Translation API. All columns from the original dataset are retained.
A handful of inaccuracies arising out of translation have been fixed so far. Therefore, the dataset can be a bit noisy. This is particularly true for coding related questions and answers. Moreover, some non-Bengali characters can be found in the text. In addition, potential duplicates from the original dataset are retained as well.