Datasets:

LVSTCK
/

winogrande-mk

This dataset is a Macedonian adaptation of the WinoGrande dataset, originally curated (English -> Serbian) by Aleksa Gordić. It was translated from Serbian to Macedonian using the Google Translate API.

You can find this dataset as part of the macedonian-llm-eval GitHub and HuggingFace.

Why Translate from Serbian?

The Serbian dataset was selected as the source instead of English because Serbian and Macedonian are closer from a linguistic standpoint, making Serbian a better starting point for translation. Additionally, the Serbian dataset was refined using GPT-4, which, according to the original report, significantly improved the quality of the translation. Note that this is an assumption that needs further validation (quantitative).. a small quality check was conducted on the translated Macedonian dataset, and the results were deemed to be of good quality.

Overview

Language: Macedonian
Source: Serbian LLM Evaluation Dataset v1
Translation: Serbian -> Macedonian (via Google Translate API)
Split: Only the test split is available.
Purpose: Benchmark and evaluate large language models in Macedonian.

Credits

Serbian version of the dataset: Aleksa Gordić
Translation: Google Translate API
Original dataset - WinoGrande Paper: WinoGrande: An Adversarial Winograd Schema Challenge at Scale

Limitations

Translation may contain minor inaccuracies.
Only the test split is provided.

Citation

If you use this dataset, please cite the original WinoGrande dataset:

@InProceedings{ai2:winogrande,
title = {WinoGrande: An Adversarial Winograd Schema Challenge at Scale},
authors={Keisuke, Sakaguchi and Ronan, Le Bras and Chandra, Bhagavatula and Yejin, Choi
},
year={2019}
}