MGEN
收藏资源简介:
MGEN是一个包含超过410万条自然出现的泛化和量化句子的数据集,这些句子从各种文本来源中提取。数据集中的句子具有长的上下文文档,对应于网站和学术论文,涵盖了11种不同的量化词。MGEN是最大的、最多样化的自然出现的泛化句子数据集,为大型计算研究打开了大门。数据集是通过一个两步流程挖掘的:一个句法过滤器检测裸复数(这是泛化最常见的语法),然后一个二元分类器将它们标记为泛化或非泛化。该流程应用于ZYDA数据集的一个子集,以收集一个多样化和准确的泛化数据集。数据集的句子长度较长,平均超过16个单词。分析数据集中的词频,我们发现说话者最常使用泛化来概括关于不同种类的内容。
MGEN is a dataset comprising over 4.1 million naturally occurring generic and quantified sentences extracted from a wide range of textual sources. Each sentence in the dataset is associated with long contextual documents from websites and academic papers, and the dataset encompasses 11 distinct quantifiers. MGEN stands as the largest and most diverse dataset of naturally occurring generic sentences to date, unlocking opportunities for large-scale computational research. The dataset was constructed via a two-step mining pipeline: first, a syntactic filter identifies bare plurals, the most prevalent grammatical structure for generic expressions, then a binary classifier labels these sentences as either generic or non-generic. This pipeline was applied to a subset of the ZYDA dataset to curate a diverse and high-quality generic sentence dataset. The sentences in this dataset are relatively long, with an average length exceeding 16 words. An analysis of word frequencies within the dataset shows that speakers most commonly use generic statements to generalize across various categories of content.

- 1通过爱丁堡大学信息学院 · 2025年



