roemmele/InSentive

InSentive数据集是用于研究自动生成文本如何影响创意写作的句子填充任务的人类创作数据。数据集分为两部分：创作块（authoring_blocks）和评判组响应（judgment_group_responses）。创作块包含与单个提示相关的、按难度等级（简单或困难）分类的创作单元，每个单元包括参与者观察生成示例前后编写的两个句子，以及他们观察到的生成示例。评判组响应包含评估任务中对单个评判组的评分者反馈，每个反馈表明评分者认为组内的哪个句子最具故事性。

The InSentive dataset is human-authored data for a sentence infilling task used to study the influence of automatically generated text on creative writing. The dataset is divided into two parts: authoring blocks and judgment group responses. Authoring blocks contain creative units associated with a single prompt categorized by difficulty level (easy or hard), each unit includes two sentences authored by participants before and after observing the generated examples, as well as the observed generated examples themselves. Judgment group responses contain raters feedback on individual judgment groups in the evaluation task, with each feedback indicating which sentence in the group the rater considers the most story-like.