RuCoLA

RuCoLA是首个大规模俄语语言可接受性分类语料库，由俄罗斯高等经济学院等机构创建。该数据集包含13400个句子，分为两部分：9800个来自语言学出版物的句子，以及3600个由生成模型产生的句子。RuCoLA旨在通过提供详细的可接受性判断和错误分类，帮助评估和改进语言模型的语法知识，特别是在处理形态和语义错误方面。此外，数据集还包括一个公共排行榜，用于评估语言模型在俄语上的表现。

RuCoLA is the first large-scale Russian language acceptability classification corpus, developed by institutions including the Higher School of Economics in Russia and other relevant organizations. This corpus contains 13,400 sentences divided into two subsets: 9,800 sentences sourced from linguistic publications, and 3,600 sentences generated by generative models. RuCoLA is designed to help evaluate and enhance the grammatical knowledge of language models, especially in handling morphological and semantic errors, by providing detailed acceptability judgments and error categorizations. Furthermore, the corpus includes a public leaderboard for assessing the performance of language models on Russian language tasks.