PASTEL

PASTEL是一个大规模的人工标注平行文体语言数据集，包含约41,000个平行句子和8,300个平行故事，每个故事平均由2.63个标注者完成。数据集涵盖了性别、年龄、民族、居住国家、教育水平、政治观点和写作时间等七种不同的人格特质。数据集通过亚马逊Mechanical Turk平台收集，确保了输入含义的准确性和标注者的文体多样性。PASTEL数据集主要用于文体语言变化的研究，特别是在控制文体分类任务和监督文体转移任务中表现出色，有助于设计更精确的实验和评估模型。

PASTEL is a large-scale manually annotated parallel stylistic language dataset. It contains approximately 41,000 parallel sentences and 8,300 parallel stories, with each story being annotated by an average of 2.63 annotators. The dataset covers seven distinct attributes including gender, age, ethnicity, country of residence, educational level, political views, and writing time. Collected via the Amazon Mechanical Turk platform, the dataset ensures the accuracy of input semantics and the stylistic diversity of annotators. The PASTEL dataset is primarily utilized for research on stylistic language variation, and delivers outstanding performance especially in controlled stylistic classification tasks and supervised stylistic transfer tasks, facilitating the design of more precise experiments and model evaluation.