schneiderkamplab/Edda-Beta

Edda是一个公开的基准数据集，旨在评估模型在丹麦语对话中检测讽刺（反语）的任务。每个例子包含一个丹麦语句子或短文、一个指示文本是否讽刺的二进制标签，以及一个人工编写的解释说明为什么标注者做出该决定。数据集支持两种评估场景：分类和解释生成/检索。这些任务对于情感分析、内容审核和需要理解丹麦语微妙之处的对话代理等实际应用非常重要。数据集面临的挑战包括丹麦讽刺的微妙性、讽刺与反语的界限模糊以及文本长度的可变性。

Edda is a publicly benchmark designed to evaluate models on the task of detecting sarcasm (irony) in Danish conversations. Each example consists of a Danish sentence or short passage, a binary label indicating whether the text is sarcastic, and a human‑written rationale that explains why the annotator reached that decision. The dataset therefore enables two complementary evaluation scenarios: Classification and Rationale generation / retrieval. Both tasks are important for practical applications such as sentiment analysis, content moderation, and conversational agents that need to interpret nuanced Danish language. The dataset faces challenges including the subtlety of Danish sarcasm, the fuzzy boundary between sarcasm and irony, and the variability in text length.