agentlans/tatoeba-paraphrases

该数据集是一个多语言的释义数据集，基于Tatoeba网站的翻译对句子构建。通过将翻译对句子链接成一个大型无向图，提取出具有相似意义的句子作为释义对。数据集以GZipped JSON文件格式提供，每个文件包含一个语言的释义句子簇。

This is a large multilingual paraphrase dataset derived from Tatoeba. If two sentences translate to the same sentence in a foreign language, then those two sentences could be paraphrases. The dataset is created by linking translation pairs of sentences in a large undirected graph and identifying connected components as paraphrases. It includes various languages with at least 10 clusters, and the data is stored in GZipped JSON files. Each line contains one cluster of paraphrase sentences in JSON format.