scifact

SciFact, a dataset of 1.4K expert-written scientific claims paired with evidence-containing abstracts, and annotated \with labels and rationales. This version differs from `allenai/scifact` on HF because we do not have separate splits \for claims and a corpus, instead we combine documents with claims that it supports or refutes, note that there are \also some documents that do not have any claims associated with them as well as there are some claims that do not \have any evidence. In the latter case we assign all such claims to the DUMMY document with ID -1 and without any text \(i.e. abstract sentences).

SciFact，一项包含1.4K条由专家撰写的科学断言及其包含证据的摘要数据集，并附带标签和理由说明。本版本与HF上的`allenai/scifact`存在差异，因为我们没有为断言和语料库提供单独的拆分，而是将支持或反驳断言的文档与断言合并。值得注意的是，也存在一些未关联任何断言的文档，以及一些没有证据支持的断言。在后一种情况下，我们将所有此类断言分配给具有ID -1的DUMMY文档，且不包含任何文本（即摘要句子）。