PASTED

该数据集名为PASTED，包含了分布内的训练集、验证集和测试集，以及一个用于检测AI改写的文本段的泛化测试集。数据集中既有由人类编写的文本，也有机器生成的文本，涵盖了多种改写风格。此外，该数据集不仅包括无上下文信息的改写，也包括了上下文相关的改写。数据被分为训练集、验证集和测试集，比例分别为80%、10%和10%。在分布内部分，数据集包含了83,089个实例（包括28,473个原文和54,616个改写文本）；在分布外部分，有9,372个实例。该数据集的任务是检测改写的文本段。

This dataset is named PASTED. It comprises in-distribution training, validation, and test sets, alongside a generalized test set designed for detecting AI-paraphrased text segments. The dataset contains both human-written and machine-generated text, covering a wide range of paraphrasing styles. Additionally, it includes both context-free paraphrases and context-dependent paraphrases. The data is split into training, validation, and test sets with a ratio of 80%, 10%, and 10% respectively. For the in-distribution portion, the dataset contains 83,089 instances in total, including 28,473 original texts and 54,616 paraphrased texts; for the out-of-distribution portion, there are 9,372 instances. The core task of this dataset is to detect paraphrased text segments.