LogicNLG

该数据集名为LogicNLG，它基于TabFact数据集构建，专注于生成自然语言陈述，这些陈述逻辑上源自开放领域半结构化表格中的事实。每个表格包含5个不同的示例，覆盖了各种类型的逻辑推理。该数据集的特点是逻辑推理丰富，主要由平均长度为11个词的短句组成，将语言复杂性隔离，以便专注于逻辑推理。其规模包括28,450个训练样本、4,260个验证样本和4,305个测试样本，这些样本基于7,392个开放领域表格。该数据集的任务是从具有逻辑蕴含的开放领域表格中进行自然语言生成。

This dataset, named LogicNLG, is built upon the TabFact dataset and focuses on generating natural language statements that are logically entailed by facts from open-domain semi-structured tables. Each table contains 5 distinct examples covering various types of logical reasoning. Characterized by rich logical reasoning, the dataset primarily consists of short sentences with an average length of 11 words, isolating linguistic complexity to focus exclusively on logical reasoning. In terms of scale, it includes 28,450 training samples, 4,260 validation samples, and 4,305 test samples, which are derived from 7,392 open-domain tables. The task of this dataset is natural language generation of statements logically entailed by facts in open-domain semi-structured tables.