VoxEL

This dataset has manual annotations with respect to Wikipedia over the same text written in five languages: German (de), English (en), Spanish (es), French (fr) and Italian (it). The dataset is composed of 15 annotated news articles (in each of the 5 languages; 75 articles in total) where there is the same number of sentences in each language, as well as the same set of annotations for each corresponding sentence in the different languages. Each language has a total of 94 sentences across the 15 articles.<br>We propose two annotated versions of the dataset: a strict version that only annotates persons, organizations and places (per, for example, traditional NER/MUC definitions of an entity), and a relaxed version that includes a larger number of annotations (e.g., capturing entity mentions such as “inflation” that have a corresponding Wikipedia article). Both the relaxed and the strict versions have the same text in the same languages. The strict version has 204 annotations per language, while the relaxed version has 674 annotations per language.<br>If you use VoxEL in a research work, we would ask you to reference the following paper that describes the dataset in detail:Henry Rosales-Méndez, Aidan Hogan, Barbara Poblete. "VoxEL: A Benchmark Dataset for Multilingual Entity Linking". International Semantic Web Conference (ISWC), Monterey, United States, 2018. [http://aidanhogan.com/docs/voxel-entity-linking.pdf]本数据集针对德语（de）、英语（en）、西班牙语（es）、法语（fr）及意大利语（it）五种语言的同一文本，提供基于维基百科（Wikipedia）的人工标注。 该数据集由15篇经人工标注的新闻文章组成（每种语言各15篇，总计75篇）。所有语言版本的句子总数完全一致，且不同语言中对应句子的标注集合也完全相同。15篇文章中，每种语言总计包含94个句子。 本数据集提供两种标注版本：严格版仅标注人物、组织与地点，符合传统命名实体识别（Named Entity Recognition, NER）与MUC（Message Understanding Conference）的实体定义；宽松版则包含更多标注内容，例如捕捉带有对应维基百科条目的“通货膨胀”这类实体提及。严格版与宽松版均使用相同语言的同一文本。严格版每种语言包含204个标注，宽松版则为674个标注。 若您在研究工作中使用VoxEL数据集，请引用如下详细描述该数据集的论文：Henry Rosales-Méndez、Aidan Hogan、Barbara Poblete。《VoxEL：面向多语言实体链接的基准数据集》，国际语义网会议（International Semantic Web Conference, ISWC），美国蒙特雷，2018年。[http://aidanhogan.com/docs/voxel-entity-linking.pdf]