Datasetpublic

Yunika/xquad-nepali

XQuAD-NP: Nepali Translation of XQuAD Overview XQuAD-NP is a Nepali translation of the XQuAD dataset, a multilingual benchmark for evaluating question-answering models. The dataset consists of Nepali translations of the original English passages, questions, and answers from the XQuAD dataset. This resource is useful for benchmarking and improving the performance of machine learning models on extractive question-answering tasks in Nepali. Dataset… See the full description on the dataset page: https://huggingface.co/datasets/Yunika/xquad-nepali.

sourceHugging Faceccupdated 1y agoView on Hugging Face

0likes18downloads

Dataset Card

XQuAD-NP: Nepali Translation of XQuAD

Overview

XQuAD-NP is a Nepali translation of the XQuAD dataset, a multilingual benchmark for evaluating question-answering models. The dataset consists of Nepali translations of the original English passages, questions, and answers from the XQuAD dataset. This resource is useful for benchmarking and improving the performance of machine learning models on extractive question-answering tasks in Nepali.

Dataset Details

—Language: Nepali (ne)
—Source: Translated from XQuAD-English
—Data Format: JSON following the SQuAD-style format
—Task: Extractive Question Answering

Data Collection Process

The dataset was created by translating the XQuAD English dataset into Nepali using a combination of automated translation and human verification. The key steps were:

1.Automated Translation of Context and Question: Initial translations of context and question were generated using Google Translate.
2.Human Review: Native Nepali speakers reviewed the translations for accuracy.
3.Answer Span Extraction Post Translation: Translation-invariant tokens (**) were employed to retain maximum answer spans post translation.
4.Quality Assurance: Sample-based evaluations were conducted to verify consistency and coherence.

Data Structure

The dataset follows the same format as XQuAD and SQuAD 1.1, with each entry containing:

—context: A passage from a Wikipedia article
—question: A corresponding question
—answers: A list of possible correct answers with their positions in the passage

Example data point:

json

  {
      "context": "प्यान्थर्सको डिफेन्सले केवल ३०८ अंक गुमायो, लिगमा छैटौं स्थानमा रह्यो, जबकि २४ इन्टरसेप्सनका साथ NFL को नेतृत्व गर्‍यो र चार प्रो बाउल छनोटहरूमा पनि अग्रता लियो। प्रो बाउल डिफेन्सिभ ट्याकल कवान शर्टले ११ स्याकसहित टोलीको नेतृत्व गरे, जबकि तीन फम्बलहरू फोर्स गरे र दुई पुन: प्राप्त गरे। साथी लाइनम्यान मारियो एडिसनले ६½ स्याक थपे। प्यान्थर्स लाइनमा अनुभवी डिफेन्सिभ एन्ड जेरेड एलेन पनि थिए, जो ५ पटकका प्रो बलर थिए जसले १३६ सहित NFL का सक्रिय करियर स्याक लिडर थिए, साथै डिफेन्सिभ एन्ड कोनी एली पनि थिए, जसले केवल ९ सुरुवातमा ५ स्याकहरू लिएका थिए। तिनीहरूको पछाडि, प्यान्थर्सका तीन सुरुवाती लाइनब्याकरहरू मध्ये दुईलाई पनि प्रो बाउलमा खेल्न छनोट गरिएको थियो: थोमस डेभिस र लुक कुएचली। डेभिसले ५½ स्याक, चार फोर्स्ड फम्बल र चार इन्टरसेप्सनहरू संकलन गरे, जबकि कुएचलीले ट्याकलहरूमा टोलीको नेतृत्व गरे (११८) दुई फम्बलहरू फोर्स गरे, र आफ्नै चार पासहरू इन्टरसेप्ट गरे। क्यारोलिनाको सेकेन्डरीमा प्रो बाउल सेफ्टी कर्ट कोलम्यान थिए, जसले करियरको उच्चतम सात इन्टरसेप्सनका साथ टोलीको नेतृत्व गरे, जबकि ८८ ट्याकलहरू पनि बनाए र प्रो बाउल कर्नरब्याक जोश नर्मन, जो सिजनको समयमा शटडाउन कर्नरमा विकसित भए र चार इन्टरसेप्सन गरे, जसमध्ये दुई टचडाउनको लागि फिर्ता गरियो।",
      "answers": {
          "text": [
              "३०८"
          ],
          "answer_start": [
              28
          ]
      },
      "question": "प्यान्थर्सको रक्षापंक्तिले कति अंक आत्मसमर्पण गर्‍यो?",
      "id": 0
  },

Uses and Applications

—Benchmarking: Evaluating multilingual and Nepali-language QA models.
—Cross-lingual Studies: Comparing the performance of models across languages.

Citation Information

@article{Yunika Bajracharya_Suban Shrestha_Saurav Bastola_Sanjivan Satyal_2025, 
  title={Extractive Nepali Question Answering System}, 
  volume={9}, 
  url={https://www.nepjol.info/index.php/kjse/article/view/78368}, 
  DOI={10.3126/kjse.v9i1.78368}, 
  number={1}, 
  journal={KEC Journal of Science and Engineering}, 
  author={Yunika Bajracharya and Suban Shrestha and Saurav Bastola and Sanjivan Satyal}, 
  year={2025}, 
  month={May}, 
  pages={95–102} 
}