E2E NLG Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/sebastianGehrmann/diverse_ensembling
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了5万个基于对话行为的意义表示（MRs）以及在餐厅领域的参考对示例。数据集被划分为76%的训练数据、9%的验证数据和15%的测试数据。在验证集中，每个MR平均有8.1个参考，并且还有一个包含630个未见MR的独立测试集。该数据集的规模为5万个示例，任务是对话到文本的数据生成。
This dataset includes 50,000 meaning representations (MRs) based on dialogue acts along with reference pair examples in the restaurant domain. It is split into three subsets: 76% for training, 9% for validation, and 15% for testing. In the validation subset, each MR averages 8.1 references, and an independent test set containing 630 unseen MRs is also included. The dataset has a total of 50,000 examples, and its core task is dialogue-to-text data generation.
提供机构：
E2E NLG challenge organizers
