Sinhala Physical Common Sense Reasoning Dataset
收藏资源简介:
该数据集是首个僧伽罗语物理常识推理数据集,由莫拉图瓦大学和梅西大学联合创建,旨在解决低资源语言僧伽罗语在自然语言处理中的研究瓶颈。数据集包含110条人工创建并验证的样本,每条样本包含提示、正确答案和错误答案,内容涵盖佛教、文学、神话、体育、食品等多个领域。数据创建过程严格遵循Global PIQA标准,由两位长期生活在斯里兰卡的僧伽罗语专家手工编写。该数据集主要用于评估语言模型在跨文化物理常识推理任务中的表现,特别关注斯里兰卡本土文化语境下的语言理解能力。
This dataset is the first Sinhala physical commonsense reasoning dataset, jointly developed by the University of Moratuwa and Massey University. It is designed to address the research bottlenecks faced by Sinhala, a low-resource language, in natural language processing (NLP) research. The dataset consists of 110 manually created and validated samples, each containing a prompt, a correct answer, and an incorrect answer. The samples cover diverse domains including Buddhism, literature, mythology, sports, food, and more. The entire dataset creation process strictly adheres to the Global PIQA standards, and was manually written by two Sinhala language experts who have long resided in Sri Lanka. This dataset is primarily intended to evaluate the performance of language models on cross-cultural physical commonsense reasoning tasks, with a special focus on language comprehension abilities within the context of Sri Lankan native culture.




