Datasets:
license: cc-by-sa-4.0
task_categories:
- question-answering
language:
- sr
tags:
- text
- extractive-qa
- generative-qa
Serbian Question-Answering Datasets
This repository provides multiple QA datasets in Serbian, suitable for training LLMs to answer questions, perform tasks, or function as chatbots.
Datasets Overview
- SQuAD-sr-md – Manually corrected subset of SQuAD-sr (~7k corrected samples), for higher reliability and accuracy.
- SerbianQA-Gen – Synthetic QA dataset (~74k samples) generated from encyclopedia articles, Wikipedia pages, and scientific abstracts. Organized into four sub-datasets:
- PaSaz
- Wikipedia
- Sveznanje
- Biblisha
Folder Structure
QuestionAnswering/
│
├─ README.md
├─ SQuAD-sr-md/
│ └─ train-00000-of-00001.parquet
├─ SerbianQA-Gen/
│ ├─ PaSaz/
│ │ └─ train-00000-of-00001.parquet
│ ├─ Wikipedia/
│ │ └─ train-00000-of-00001.parquet
│ ├─ Sveznanje/
│ │ └─ train-00000-of-00001.parquet
│ └─ Biblisha/
│ └─ train-00000-of-00001.parquet
About the Work
The datasets in this repository are introduced in the following paper:
Development of Serbian QA Datasets through Prompt-Based Generation and Human Validation
Jovana Rađenović, Olivera Kitanović, Ranka Stanković, Mihailo Škorić
Faculty of Mining and Geology, University of Belgrade, Serbia
{jovana.radjenovic, olivera.kitanovic, ranka.stankovic, mihailo.skoric}@rgf.bg.ac.rs
Status: Accepted, not yet published
This work presents the development of two question–answering (QA) datasets for the Serbian language: SQuAD-sr-md and SerbianQA-Gen.
The SQuAD-sr-md dataset contains around 7k examples and represents a manually refined subset of the original SQuAD-sr dataset.While SQuAD-sr is a significant resource for Serbian QA research, it was created through a translation-based approach using an adapted Translate–Align–Retrieve method, which introduced various linguistic inconsistencies. The manual revision process, although time-consuming, substantially improved the grammatical correctness, terminological consistency, and overall reliability of the dataset, making it more suitable for training extractive QA models. We fine-tuned a several Serbian base encoder models on (i) our SQuAD-sr-md dataset and (ii) a subset of SQuAD-sr that corresponds exactly to the articles and QA pairs thatwere refined to create the SQuAD-sr-md. Evaluation of these models highlighted the TeslaXLM base model as a particularly promising candidate for future extractive QA training. The results also indicate that the refinement of the dataset positively impacts model performance.
SQuAD-sr subset
| Base model | EM (%) | F1 (%) |
|---|---|---|
| BERTić | 71 | 81.29 |
| Jerteh-81 | 48 | 57.33 |
| Jerteh-355 | 64 | 71.64 |
| XLM-r-BERTić | 79 | 87.94 |
| XLMali | 68 | 78.74 |
| TeslaXLM | 80 | 89.10 |
| Average | 68.83 | 77.75 |
SQuAD-sr-md
| Base model | EM (%) | F1 (%) |
|---|---|---|
| BERTić | 71 | 80.61 ↓ |
| Jerteh-81 | 52 ↑ | 61.55 ↑ |
| Jerteh-355 | 60 ↓ | 73.01 ↑ |
| XLM-r-BERTić | 77 ↓ | 88.49 ↑ |
| XLMali | 75 ↑ | 83.00 ↑ |
| TeslaXLM | 83 ↑ | 92.23 ↑ |
| Average | 69.67 ↑ | 79.81 ↑ |
↑ indicates improvement compared to the original SQuAD-sr subset, while ↓ indicates a decrease.
The SerbianQA-Gen dataset consists of approximately 74k automatically generated QA pairs produced with the assistance of large language models (LLMs). The dataset covers a wide range of topics and sources and is currently undergoing additional manual revision to improve grammatical accuracy, fluency, and naturalness.
Together, these datasets contribute to addressing the lack of large-scale QA resources for the Serbian language. They enable the training and evaluation of language models capable of deeper semantic understanding of Serbian, a language characterized by rich morphology and complex syntax. By making these datasets publicly available, we aim to support further research in Serbian NLP and encourage the development of more robust QA and conversational AI systems for the language.
Citation
@unpublished{SerbianQA2026,
title={Development of Serbian QA Datasets through Prompt-Based Generation and Human Validation},
author={Rađenović, Jovana and Kitanović, Olivera and Stanković, Ranka and Škorić, Mihailo},
year={2026},
note={Accepted, not yet published}
}
|
Истраживање jе спроведено уз подршку Фонда за науку Републике Србиjе, #7276, Text Embeddings – Serbian Language Applications – TESLA |
This research was supported by the Science Fund of the Republic of Serbia, #7276, Text Embeddings - Serbian Language Applications - TESLA |