README.md

metadata

license: apache-2.0

Dataset Card for SQuAD 1.1 Nepali Translation

Dataset Summary

This dataset is a translated version of the Stanford Question Answering Dataset (SQuAD) 1.1 into Nepali. SQuAD 1.1 is a widely used benchmark dataset for extractive question-answering tasks, originally developed in English. By translating it into Nepali, we aim to provide a high-quality dataset for evaluating and improving NLP models for Nepali, an underrepresented language in the field of natural language processing.

Dataset Creation

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 with LLM: GPT-4 was used as an additional check to verify consistency and coherence.

Source Data

• Original Dataset: Stanford Question Answering Dataset (SQuAD) 1.1. SQuAD 1.1
• Translation Method: The dataset was translated manually and/or using machine translation tools, followed by human verification to ensure accuracy and linguistic fidelity.

Intended Use

The dataset is intended for research and development in:

• Extractive question-answering systems for Nepali
• Multilingual and low-resource NLP research
• Fine-tuning and evaluation of large language models

Languages

• Primary Language: Nepali
• Script: Devanagari