asseco-group/incoherence-bench

Polish Document Incoherence Benchmark是一个针对波兰语文档级不一致性/连贯性检测的基准数据集，被定义为一种类似于NLI的分类任务，具有针对审计程序、法律和组织文本的适应标签语义。该数据集仅用于评估，包含一个测试集。数据集的特征包括前提、假设和标签，标签分为蕴含、中立和矛盾三种类型。

Polish Document Incoherence Benchmark is a benchmark dataset for document-level inconsistency/coherence detection in Polish, formulated as an NLI-like classification task with adapted label semantics for auditing procedural, legal, and organizational texts. This dataset is for evaluation-only and includes a single split: test. The features of the dataset include premise, hypothesis, and label, with the label being categorized into entailment, neutral, and contradiction.