metadata

language:
  - sr
license: cc-by-sa-4.0
pretty_name: Serbian SQuAD
size_categories:
  - 100K<n<1M
source_datasets:
  - rajpurkar/squad
task_categories:
  - question-answering
task_ids:
  - extractive-qa

Dataset Card for Serbian SQuAD

Dataset Description

Repository: Hugging Face Dataset
Point of Contact: [SmartCat]
Original Data: https://huggingface.co/datasets/rajpurkar/squad
** Original data in Serbian: https://www.kaggle.com/datasets/aleksacvetanovic/squad-sr

Dataset Summary

This dataset is an automatic Serbian translation of the Stanford Question Answering Dataset (SQuAD) 1.1. The original SQuAD, developed by Stanford, is a reading comprehension dataset consisting of questions posed by crowdworkers on a set of Wikipedia articles. The answer to every question is a segment of text (span) from the corresponding reading passage, or the question might be unanswerable. It's the largest Serbian QA dataset with more than 87K samples in both Cyrillic and Latin.

The translation was performed automatically using the GPT-3.5-turbo-0125 model, preserving the structure and format of the original dataset while providing content in Serbian.

Supported Tasks and Leaderboards

Extractive Question Answering: The dataset can be used to train and evaluate models for extractive question answering tasks in Serbian.

Languages

The dataset is in Serbian (sr). The original dataset was in English (en).

Dataset Structure

Data Instances

Each instance in the dataset contains:

A passage of text from a Wikipedia article (context)
A question about that passage
The answer to the question, which is a span of text from the passage

Data Fields

id: A string identifier for the example
title: The title of the Wikipedia article
context: The passage of text (paragraph from Wikipedia)
question: The question posed about the context

Dataset Creation

Curation Rationale

This dataset was created to provide a resource for Serbian language question answering tasks, leveraging the high-quality structure and content of the original SQuAD dataset.

Source Data

Initial Data Collection and Normalization

The source data is the SQuAD 1.1 dataset, which contains 100,000+ question-answer pairs on 500+ articles. The original data was retrieved from: https://huggingface.co/datasets/rajpurkar/squad

Who are the source language producers?

The original questions were posed by crowdworkers on English Wikipedia articles. The answers were extracted from these articles.

Annotations

Annotation process

The original English dataset was automatically translated to Serbian using the GPT-3.5-turbo-0125 model.

Who are the annotators?

The translation was performed automatically by an AI model, without human intervention. The original annotations were created by crowdworkers.

Personal and Sensitive Information

The dataset is based on Wikipedia articles and should not contain personal or sensitive information. However, as it's an automatic translation of web content, users should be aware of potential inaccuracies or unintended content.

Considerations for Using the Data

Social Impact of Dataset

This dataset contributes to the development of question answering systems for the Serbian language, potentially improving access to information and language technologies for Serbian speakers.

Discussion of Biases

The dataset may inherit biases present in the original SQuAD dataset, including biases in Wikipedia content and in the questions posed by crowdworkers. Additionally, the automatic translation process may introduce its own biases or errors.

Other Known Limitations

The quality of the Serbian translations has not been manually verified and may contain errors.
The automatic translation process may not capture nuances or context-specific meanings from the original text.
The dataset maintains the same limitations as the original SQuAD 1.1, such as the potential for artificial questions that might not reflect real-world information-seeking behavior.

Additional Information

Licensing Information

This dataset is licensed under CC-BY-SA-4.0, following the licensing of the original SQuAD dataset.

Citation Information

If you use this dataset, please cite both the original SQuAD dataset and this Serbian translation:



@article

{rajpurkar2016squad,
title={Squad: 100,000+ questions for machine comprehension of text},
author={Rajpurkar, Pranav and Zhang, Jian and Lopyrev, Konstantin and Liang, Percy},
journal={arXiv preprint arXiv:1606.05250},
year={2016}
}



@misc

{cvetanović2024syntheticdatasetcreationfinetuning,
title={Synthetic Dataset Creation and Fine-Tuning of Transformer Models for Question Answering in Serbian},
author={Aleksa Cvetanović and Predrag Tadić},
year={2024},
eprint={2404.08617},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2404.08617},
}



@misc

{serbian-squad,
title={Serbian SQuAD Dataset},
author={[Your Name]},
year={2024},
howpublished={\url{https://huggingface.co/datasets/your-username/serbian-squad}}
}

Contributions

Thanks to Stanford for creating the original SQuAD dataset. Thank you to Aleksa Cvetanovic and Predrag Tadic for translating SQuAD to Serbian and uploading the dataset to Kaggle.

Loading the Dataset

Here's a Python code example to load the dataset using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("smartcat/squad_sr")