CAMeL-Lab/BAREC-Shared-Task-2025-doc

BAREC 2025是一个面向细粒度阿拉伯语可读性评估的大规模数据集，包含超过1百万个单词，覆盖19个可读性级别，并可以映射到7级、5级和3级的更粗略方案。数据集在句子级别进行标注，并根据文档中最难句子的可读性级别来确定文档级别的可读性分数。该数据集支持多类可读性分类任务，并包含训练、验证和测试数据集。

BAREC 2025 is a large-scale dataset for fine-grained Arabic readability assessment, containing over 1 million words annotated across 19 readability levels, and can be mapped to coarser schemes of 7, 5, and 3 levels. The dataset is annotated at the sentence level, and document-level readability scores are determined by the readability level of the most difficult sentence in the document. The dataset supports multi-class readability classification tasks and includes training, development, and test sets.