absinth

该数据集名为“absinth”，包含了德国新闻文章及其生成的摘要，这些摘要由人工标注以检测幻觉现象。该数据集由4,314个文章-摘要句子对组成，这些句子对被标注为忠实、内在或外在的幻觉。此外，该数据集还包括了12位母语为德语的标注者进行的手动标注，确保了高质量标注和标注者之间的一致性。规模上，数据集包含了4,314个文章-摘要对，其任务集中在新闻摘要中的幻觉检测。

This dataset is named 'absinth'. It contains German news articles and their generated summaries, which are manually annotated for hallucination detection. The dataset comprises 4,314 article-summary sentence pairs, with these pairs annotated as faithful, intrinsic hallucination, or extrinsic hallucination. Additionally, the dataset includes manual annotations conducted by 12 native German speakers, ensuring high-quality annotations and inter-annotator agreement. With 4,314 article-summary pairs in total, the core task of this dataset is hallucination detection in news summarization.