GINCO Training Dataset

GINCO Training Dataset是由斯洛文尼亚约瑟夫·斯特凡研究所创建的一个用于自动体裁识别的训练数据集，包含1,125个斯洛文尼亚网络文档，总计约65万字。该数据集通过新的标注方案进行手动标注，旨在提高标签清晰度和标注者间一致性。数据集涵盖了网络数据相关的多种挑战，如机器翻译内容、编码错误等，使得分类器能在真实条件下进行评估。该数据集主要用于解决网络文档的体裁识别问题，支持深入分析新提供语料库的质量和组成，同时也有助于提升语言技术在多个领域的应用，如词性标注、零样本依赖解析、自动摘要和机器翻译等。

The GINCO Training Dataset is a training dataset for automatic genre recognition, developed by the Jožef Stefan Institute in Slovenia. It comprises 1,125 Slovenian web documents, with a total of approximately 650,000 words. This dataset was manually annotated via a novel annotation scheme, with the goal of enhancing label clarity and inter-annotator agreement. The dataset encompasses various challenges inherent in web data, such as machine-translated content and encoding errors, allowing classifiers to be evaluated under real-world conditions. Primarily intended to solve the genre recognition task for web documents, the dataset supports in-depth analyses of the quality and composition of the newly released corpus, and also contributes to advancing language technology applications across multiple domains, including part-of-speech tagging, zero-shot dependency parsing, automatic summarization, and machine translation.