ronantakizawa/jfleg-ja

JFLEG-JA是一个日语语法错误修正（GEC）数据集，由原始的JFLEG（JHU FLuency-Extended GUG）基准启发创建。该数据集包含1335个含有语法错误的日语句子，每个句子都伴有4个人类高质量修正版本，这些修正既关注语法正确性也关注流畅性。数据集涵盖多种日语语法错误类型，如助词错误、动词变形错误、语体错误、词汇选择错误、数量词错误、句子结构错误、连词错误、修饰语错误、助动词错误和条件句错误等。数据集是使用AI生成的句子和手动编写的替换句子创建的，并经过严格的质量检查，确保每个例子都有4个独特的修正版本，没有重复的句子或修正，且错误句子看起来自然和真实，修正后的句子听起来像母语者所说。

JFLEG-JA is a Japanese grammatical error correction (GEC) dataset inspired by the original JFLEG (JHU FLuency-Extended GUG) benchmark. The dataset contains 1,335 Japanese sentences with grammatical errors, each accompanied by 4 human-quality corrections focusing on both grammaticality and fluency. It covers various types of Japanese grammatical errors, including particle errors, verb conjugation errors, register/formality errors, word choice errors, counter word errors, sentence structure errors, conjunction errors, modifier errors, auxiliary verb errors, and conditional errors. The dataset was created using a combination of AI-generated sentences and manually written replacement sentences, and has undergone rigorous quality checking to ensure natural and realistic errors and native-sounding corrections.