SwaLUE
收藏资源简介:
SwaLUE是一个旨在评估斯瓦希里语自然语言处理模型性能的基准数据集,由汉阳大学数据科学系创建。该数据集包含多种类型的问答对,涵盖阅读理解、多选题、事实问答和无法回答的问题等格式。数据来源多样,包括直接从斯瓦希里语使用者收集的初级数据、现有的斯瓦希里语数据集、众包内容以及翻译的英语数据集。数据集的创建过程注重数据隐私、偏见缓解和包容性,旨在支持机器翻译、信息检索和医疗聊天机器人等应用,促进东非地区的科技创新和低资源语言的NLP研究。
SwaLUE is a benchmark dataset developed to evaluate the performance of Swahili natural language processing (NLP) models, created by the Department of Data Science at Hanyang University. This dataset includes diverse question-answer pairs spanning multiple formats, such as reading comprehension tasks, multiple-choice questions, factual question answering, and unanswerable questions. Its data sources are varied, encompassing primary data collected directly from Swahili speakers, existing Swahili datasets, crowdsourced content, and translated English datasets. Great emphasis has been placed on data privacy, bias mitigation and inclusivity throughout the dataset's development process. This benchmark aims to support applications including machine translation, information retrieval and medical chatbots, while promoting technological innovation in East Africa and NLP research on low-resource languages.

- 1SwaQuAD-24: QA Benchmark Dataset in Swahili汉阳大学数据科学系 · 2024年



