mASNQ, mWikiQA, mTRECQA

本研究介绍了三个新的多语言答案句子选择（AS2）数据集：mASNQ、mWikiQA和mTRECQA，涵盖法语、德语、意大利语、葡萄牙语和西班牙语五种欧洲语言，总计超过1亿对问题-答案。这些数据集通过监督自动机器翻译（AMT）从现有的英语AS2数据集如ASNQ、WikiQA和TREC-QA转换而来，使用大型语言模型（LLM）。创建过程包括翻译和质量评估，旨在解决低资源语言在问答系统中的性能差距问题，为多语言AS2模型的训练提供高质量数据。

This study presents three novel multilingual answer sentence selection (AS2) datasets: mASNQ, mWikiQA, and mTRECQA, which cover five European languages including French, German, Italian, Portuguese, and Spanish, with a total of over 100 million question-answer pairs. These datasets are derived from existing English AS2 datasets such as ASNQ, WikiQA, and TREC-QA via supervised automatic machine translation (AMT) utilizing large language models (LLMs). The creation process includes translation and quality assessment, aiming to address the performance gap of low-resource languages in question answering systems and provide high-quality training data for multilingual AS2 models.