arXiv 2408.106042025-02-19cs.CLcs.AIcs.IRcs.LG

基于答案段落选择的多语言非事实型问答

Multilingual Non-Factoid Question Answering with Answer Paragraph Selection

Indraprastha Institute of Information Technology, Delhi（因德拉普拉斯信息技术学院德里校区）；International Institute of Information Technology, Hyderabad（国际信息技术学院海得拉巴校区）

Ritwik Mishra, Sreeram Vennam, Rajiv Ratn Shah, Ponnurangam Kumaraguru

备注：: Shorter version accepted into DSFA, a special session in PAKDD 2025, Sydney

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出覆盖38种语言、含578K余问答对的非事实型多语言数据集MuNfQuAD，并通过微调APS模型验证其答案段落选择与跨语言泛化能力。

摘要：

大多数现有问答数据集（QuAD）主要关注高资源语言中基于事实型的短上下文问答（QA）。然而，此类数据集在低资源语言方面的覆盖范围仍然有限，只有少数工作聚焦于基于事实型的 QuAD，而没有任何工作关注非事实型 QuAD。因此，本工作提出了 MuNfQuAD，这是一个包含非事实型问题的多语言 QuAD。该数据集利用 BBC 新闻文章中的疑问式小标题作为问题，并将相应段落作为银标准答案。数据集包含覆盖 38 种语言的超过 578K 个问答对，其中包括若干低资源语言，是迄今为止规模最大的多语言问答数据集。基于对来自 MuNfQuAD 的 790 个问答对的人工标注（黄金集），我们观察到 98% 的问题可以使用其对应的银标准答案回答。我们经过微调的答案段落选择（APS）模型优于基线模型。APS 模型在 MuNfQuAD 测试集和黄金集上分别取得了 80% 和 72% 的准确率，以及 72% 和 66% 的宏平均 F1 分数。此外，即使是在银标签上进行微调之后，APS 模型仍能有效泛化到黄金集中的某种语言。我们还观察到，经过微调的 APS 模型有助于缩减问题的上下文。这些发现表明，该资源将为问答研究界作出有价值的贡献。

英文摘要：

Most existing Question Answering Datasets (QuADs) primarily focus on factoid-based short-context Question Answering (QA) in high-resource languages. However, the scope of such datasets for low-resource languages remains limited, with only a few works centered on factoid-based QuADs and none on non-factoid QuADs. Therefore, this work presents MuNfQuAD, a multilingual QuAD with non-factoid questions. It utilizes interrogative sub-headings from BBC news articles as questions and the corresponding paragraphs as silver answers. The dataset comprises over 578K QA pairs across 38 languages, encompassing several low-resource languages, and stands as the largest multilingual QA dataset to date. Based on the manual annotations of 790 QA-pairs from MuNfQuAD (golden set), we observe that 98\% of questions can be answered using their corresponding silver answer. Our fine-tuned Answer Paragraph Selection (APS) model outperforms the baselines. The APS model attained an accuracy of 80\% and 72\%, as well as a macro F1 of 72\% and 66\%, on the MuNfQuAD testset and the golden set, respectively. Furthermore, the APS model effectively generalizes a certain language within the golden set, even after being fine-tuned on silver labels. We also observe that the fine-tuned APS model is beneficial for reducing the context of a question. These findings suggest that this resource would be a valuable contribution to the QA research community.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。