README.md

---
license: apache-2.0
tags:
- bangla
- punctuation-restoration
- text
- language-modeling
language:
- bn
pretty_name: Clean Oscar Bangla Punctuation Dataset
dataset_info:
  features:
  - name: Row Content
    dtype: string
  splits:
  - name: train
    num_bytes: 8595523313
    num_examples: 16040735
  download_size: 3248700472
  dataset_size: 8595523313
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Dataset Card for Clean Oscar Bangla Punctuation Dataset

## Dataset Summary

This dataset is a **cleaned version** of the `oscar_23_01_bn_train` dataset, curated specifically for training **Bangla sentence punctuation restoration models**. It includes cleaned and normalized Bangla text with annotations for various punctuation marks. The dataset is suitable for sequence labeling and punctuation restoration tasks in natural language processing (NLP).

## Supported Tasks

- **Punctuation Restoration**: Predict punctuation marks for Bangla text sequences.
- **Language Modeling**: Train models that understand Bangla sentence structure and punctuation usage.

## Languages

- **Bangla (bn)**

## Dataset Structure

The dataset contains sentences in Bangla with associated punctuation mark annotations. It is typically used for supervised training in models that restore punctuation in plain Bangla text.

### Punctuation Marks Distribution

The dataset includes the following punctuation types:

| Punctuation Type | Count      |
|------------------|------------|
| DARI (।)         | 44,697,345 |
| COMMA (,)        | 19,770,040 |
| QUESTION (?)     | 2,425,348  |
| HYPHEN (-)       | 5,278,590  |
| EXCLAMATION (!)  | 1,767,095  |
| COLON (:)        | 1,394,613  |
| SEMICOLON (;)    |   313,558  |

## Usage Example

Here's how to load the dataset using the 🤗 Datasets library:

```python
from datasets import load_dataset

dataset = load_dataset("abdullahalmunem/ha-pr-bn-munem-generated")
print(dataset["train"][0])