Attribute-Annotated Polish TV Dialogue Dataset

该数据集是一个包含波兰电视对话的新颖数据集，对话内容带有与说话者和对话者相关的属性标注，特别关注性别和数量。该数据集为这些属性提供了句子级别的标注，并且这些属性检测的有效性已经通过一名母语为波兰语的人士进行了验证，其检测精度和召回率均较高。数据集规模包括67.3万个带有属性的句子；开发集和测试集共有4千个独特的例子。任务的目的是进行带有属性控制的神经机器翻译。

This dataset is a novel corpus of Polish television dialogues, where each dialogue is annotated with attributes related to its speakers and interlocutors, with a particular focus on gender and participant count. The dataset provides sentence-level annotations for these attributes, and the validity of the attribute detection has been verified by a native Polish speaker, achieving both high precision and recall. The dataset contains 673,000 annotated sentences with associated attributes, while the development and test sets collectively include 4,000 unique instances. The core task supported by this dataset is attribute-controlled neural machine translation.