MINWIKISPLIT
收藏资源简介:
MINWIKISPLIT是一个由圣加仑大学创建的大型句子分割数据集,包含203,000对复杂的源句及其简化的目标句。该数据集通过将每个输入句子分解为一系列最小命题,即每个句子都是自包含的、无法进一步分解的语义单元,来简化复杂的语言结构。创建过程中,使用了35条手工编写的转换规则来分解各种语言结构,并通过依赖解析和词性标注的启发式方法来保证输出质量。该数据集主要用于训练自然语言处理模型,以改善下游语义应用的性能,特别是在处理复杂句子结构时的效率和准确性。
MINWIKISPLIT is a large-scale sentence splitting dataset developed by the University of St. Gallen, which contains 203,000 pairs of complex source sentences and their simplified target sentences. It simplifies complex linguistic structures by decomposing each input sentence into a series of minimal propositions, where each simplified sentence is a self-contained semantic unit that cannot be further decomposed. During its creation, 35 hand-crafted transformation rules were used to decompose various linguistic structures, and heuristic methods based on dependency parsing and part-of-speech tagging were adopted to ensure output quality. This dataset is primarily utilized for training natural language processing models to improve the performance of downstream semantic applications, particularly the efficiency and accuracy when handling complex sentence structures.

- 1MinWikiSplit: A Sentence Splitting Corpus with Minimal Propositions圣加仑大学 · 2019年



