BasPhyCo
收藏资源简介:
BasPhyCo是由巴斯克大学团队开发的物理常识推理数据集,包含356个五句子故事(117个合理/239个不合理),涵盖标准巴斯克语和西方方言双版本。数据源自意大利语GITA数据集的专业翻译与本地化改造,通过人工翻译和Latxa-3.1-Instruct模型自动生成方言变体。该数据集聚焦日常物理场景的叙事合理性判断,支持模型在低资源语言环境下进行三层级推理评估(合理性分类、冲突检测、物理状态验证),为研究多语言模型在方言场景下的物理常识理解能力提供基准。
BasPhyCo is a physical commonsense reasoning dataset developed by the research team at the University of the Basque Country. It comprises 356 five-sentence stories (117 reasonable, 239 unreasonable), with two variants: standard Basque and Western Basque dialect. Derived from the Italian GITA dataset via professional translation and localization adaptation, its dialect variants were generated through human translation and automatic synthesis using the Latxa-3.1-Instruct model. Focused on narrative plausibility judgment in daily physical scenarios, this dataset supports three-tier reasoning evaluation for models in low-resource language settings, namely plausibility classification, conflict detection, and physical state verification. It serves as a benchmark for investigating the physical commonsense understanding capabilities of multilingual models in dialect-based scenarios.



