JamPatoisNLI

JamPatoisNLI是首个针对克里奥尔语种Jamaican Patois的自然语言推理数据集。该数据集由斯坦福大学计算机科学系创建，旨在推动对传统服务不足语言的研究，并作为理解跨语言NLP的有用基准。数据集包含自然发生的命题和专家编写的假设，共计650个例子，分为训练、开发和验证集。JamPatoisNLI的创建过程避免了翻译偏差，增强了数据集的生态有效性，使其成为评估跨语言转移方法的更真实、相关的基准。该数据集特别适用于研究与高资源基础语言有独特关系的克里奥尔语种的跨语言转移问题。

JamPatoisNLI is the first natural language inference (NLI) dataset targeting Jamaican Patois, a creole language. It was created by the Department of Computer Science at Stanford University, with the aim of advancing research on traditionally underrepresented languages and serving as a valuable benchmark for cross-lingual NLP research. The dataset contains 650 total examples, consisting of naturally occurring premises and expert-written hypotheses, and is split into training, development, and validation sets. The creation process of JamPatoisNLI avoids translation bias, which enhances the ecological validity of the dataset, making it a more authentic and relevant benchmark for evaluating cross-lingual transfer approaches. This dataset is particularly well-suited for research on cross-lingual transfer issues in creole languages that have unique relationships with high-resource base languages.