README.md

---
dataset_info:
  features:
  - name: arabic
    dtype: string
  - name: english
    dtype: string
  splits:
  - name: train
    num_bytes: 27878710
    num_examples: 65043
  - name: test
    num_bytes: 430500
    num_examples: 1000
  download_size: 14722818
  dataset_size: 28309210
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
task_categories:
- translation
language:
- ar
- en
pretty_name: ATHAR
size_categories:
- 10K<n<100K
license: cc-by-sa-4.0
---


<!-- 
# Dataset Card for "ATHAR"

# ATHAR: A High-Quality and Diverse Dataset for Classical Arabic to English Translation

> Welcome to ATHAR Dataset

<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg" width = "150px"/>
  <p align="center"> [ <a href="https://arxiv.org/abs/2407.19835">Paper</a> ]<!-- - <a href="https://github.com/ARBML/CIDAR">GitHub</a> ]</p>

</p>


![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg) 

<div align="center">
    <picture>
        <source 
        srcset="https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg"
        media="(prefers-color-scheme: dark)"
        />
        <source
        srcset="https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg"
        media="(prefers-color-scheme: light), (prefers-color-scheme: no-preference)"
        />
        <img src="https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg" width="100%" height="350px" />
    </picture>
</div>


## Datasets Summary

The ATHAR dataset is divided into two main subsets: test data and training data.

   | Name             | Description                                                                                                             |
   |------------------|-------------------------------------------------------------------------------------------------------------------------|
   | `Training Data`  | Contains 65,043 rows of classical Arabic texts and their English translations, used for training translation models.    |
   | `Test Data`      | Contains 1,000 rows of classical Arabic texts and their English translations, reserved for evaluating model performance.|





## Dataset Structure

Each entry in the dataset is represented by a row with the following fields:

   | Field     | Description                                                                                                 |
   |-----------|-------------------------------------------------------------------------------------------------------------|
   | `Arabic(str)`  | The original text in classical Arabic.                                                                 |
   | `English(str)` | The English translation of the classical Arabic text.                                                  |




## Loading The Dataset
You can download the dataset directly from HuggingFace or use the following code:

```python
from datasets import load_dataset

athar = load_dataset("mohamed-khalil/ATHAR")
```


## Sample From The Dataset:

Below is a sample from the ATHAR dataset, showcasing a classical Arabic text and its English translation.

 **`Arabic`**: فَلم يزل الْفلس يعبد حَتَّى ظهرت دَعْوَة النَّبِي عَلَيْهِ السَّلَام فَبعث إِلَيْهِ على ابْن أَبِي طَالِبٍ فَهَدَمَهُ وَأَخَذَ سَيْفَيْنِ كَانَ الْحَارِثُ بن أبي شمرٍ الغساني, ملك غَسَّان قَلَّدَهُ إِيَّاهُمَا, يُقَالُ لَهُمَا مِخْذَمٌ وَرَسُوبٌ(وَهُمَا السَّيْفَانِ اللَّذَانِ ذَكَرَهُمَا عَلْقَمَةُ بْنُ عَبْدَةَ فِي شِعْرِهِ). فَقَدِمَ بِهِمَا عَلِيُّ بْنُ أَبِي طَالِبٍ عَلَى النَّبِيِّ صَلَّى اللَّهُ عَلَيْهِ وَسَلَّمَ فَتَقَلَّدَ أَحَدَهُمَا ثُمَّ دَفَعَهُ إِلَى عَلِيِّ بْنِ أَبِي طَالِبٍ, فَهُوَ سَيْفُهُ الَّذِي كَانَ يَتَقَلَّدُهُ

 **`English`**: Al-Fals continued to be worshipped until the advent of the Prophet, at which time ‘Ali ibn-abi-Talib was dispatched to destroy it. ‘Ali destroyed the idol and carried away therefrom two swords called Mikhdham and Rasub (the same two swords which ‘Alqamah ibn-’Abadah had mentioned in his poetry), which al-Harith ibn-abi-Shamir, king of Ghassan, had presented al-Fals. ‘Ali brought them to the Prophet who wore one of them and gave it back to him. It was the sword which ‘Ali was always wont to wear

 -->



# Dataset card

## Description

### ATHAR: A High-Quality and Diverse Dataset for Classical Arabic to English Translation

> Welcome to ATHAR Dataset

The ATHAR dataset is a comprehensive collection of classical Arabic texts translated into English. This dataset contains approximately 66,000 rows of translated texts, including the original Classical Arabic texts and their English translations.

<p align="center">
<!-- <img src="https://cdn-uploads.huggingface.co/production/uploads/64af7c627ab7586520ed8688/SwkKk3Z6kT5VZ3Oj-bVMC.jpeg" width = "150px"/> -->
  <p align="center"> [ <a href="https://arxiv.org/abs/2407.19835">Paper</a> ]<!-- - <a href="https://github.com/ARBML/CIDAR">GitHub</a> ]</p> -->

</p>



## License

The ATHAR dataset and weight diffs are licensed under **CC BY NC 4.0**
[Creative Commons NonCommercial (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/deed.en).



## Attribution

```bibtex
@misc{khalil2024atharhighqualitydiversedataset,
      title={ATHAR: A High-Quality and Diverse Dataset for Classical Arabic to English Translation}, 
      author={Mohammed Khalil and Mohammed Sabry},
      year={2024},
      eprint={2407.19835},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2407.19835}, 
}
```

## Language codes

### Arabic 

* ISO 639-3: eng
* ISO 15924: Arab
* Glottocode: arab1395	 <!-- Classical Arabic Code: clas1259 -->

### English

* ISO 639-3: ara
* ISO 15924: Latn
* Glottocode: stan1293



## Additional language information

<!-- Any relevant additional information on the language, such as:
* A list of reference publications and software (dictionaries, grammars, spellcheckers).
* If applicable, any additional information about dialectal variation that is not captured by the Glottocode.
* If relevant, the orthography used in your contribution. --> 



## Workflow

The ATHAR dataset comprises 66K sentences extracted from seminal Classical Arabic texts, expertly translated into English. These texts are cornerstone works spanning diverse genres and historical periods, providing profound insights into Islamic and world history, philosophy, science, medicine, and culture. This rich literary heritage is showcased in our dataset, sourced from the Rasaif website at [rasaif.com](https://rasaif.com/) At the time of data collection and publication of this work, there were no restrictions on scraping and using resources from this website.


## Additional guidelines

<!-- Were any additional guidelines agreed upon? 
Examples might include style guidelines, 
the use of particular grammatical forms or sentence structures, 
specific spelling or punctuation rules to be followed, etc. -->


## Datasets Summary

The ATHAR dataset is divided into two main subsets: test data and training data.

   | Name             | Description                                                                                                             |
   |------------------|-------------------------------------------------------------------------------------------------------------------------|
   | `Training Data`  | Contains 65,043 rows of classical Arabic texts and their English translations, used for training translation models.    |
   | `Test Data`      | Contains 1,000 rows of classical Arabic texts and their English translations, reserved for evaluating model performance.|





## Dataset Structure

Each entry in the dataset is represented by a row with the following fields:

   | Field     | Description                                                                                                 |
   |-----------|-------------------------------------------------------------------------------------------------------------|
   | `Arabic(str)`  | The original text in classical Arabic.                                                                 |
   | `English(str)` | The English translation of the classical Arabic text.                                                  |




## Loading The Dataset
You can download the dataset directly from HuggingFace or use the following code:

```python
from datasets import load_dataset

athar = load_dataset("mohamed-khalil/ATHAR")
```


## Sample From The Dataset:

Below is a sample from the ATHAR dataset, showcasing a classical Arabic text and its English translation.

 **`Arabic`**: فَلم يزل الْفلس يعبد حَتَّى ظهرت دَعْوَة النَّبِي عَلَيْهِ السَّلَام فَبعث إِلَيْهِ على ابْن أَبِي طَالِبٍ فَهَدَمَهُ وَأَخَذَ سَيْفَيْنِ كَانَ الْحَارِثُ بن أبي شمرٍ الغساني, ملك غَسَّان قَلَّدَهُ إِيَّاهُمَا, يُقَالُ لَهُمَا مِخْذَمٌ وَرَسُوبٌ(وَهُمَا السَّيْفَانِ اللَّذَانِ ذَكَرَهُمَا عَلْقَمَةُ بْنُ عَبْدَةَ فِي شِعْرِهِ). فَقَدِمَ بِهِمَا عَلِيُّ بْنُ أَبِي طَالِبٍ عَلَى النَّبِيِّ صَلَّى اللَّهُ عَلَيْهِ وَسَلَّمَ فَتَقَلَّدَ أَحَدَهُمَا ثُمَّ دَفَعَهُ إِلَى عَلِيِّ بْنِ أَبِي طَالِبٍ, فَهُوَ سَيْفُهُ الَّذِي كَانَ يَتَقَلَّدُهُ

 **`English`**: Al-Fals continued to be worshipped until the advent of the Prophet, at which time ‘Ali ibn-abi-Talib was dispatched to destroy it. ‘Ali destroyed the idol and carried away therefrom two swords called Mikhdham and Rasub (the same two swords which ‘Alqamah ibn-’Abadah had mentioned in his poetry), which al-Harith ibn-abi-Shamir, king of Ghassan, had presented al-Fals. ‘Ali brought them to the Prophet who wore one of them and gave it back to him. It was the sword which ‘Ali was always wont to wear