mashey/dv-synthetic-errors-lg

该数据集包含迪维希语（Dhivehi）的句子对：原始正确的句子及其合成的含错误版本。数据集总计约720万个句子对，分为训练集（80%，约580万对）、验证集（10%，约72万对）和测试集（10%，约72万对）。每个句子对包括以下字段：original（原始正确句子）、error（含合成错误的句子）和error_types（引入的错误类型）。合成错误通过基于规则的转换生成，包括变音符号错误、字符错误、时态标记错误、格标记错误、词序问题、一致性错误和后缀错误。This dataset contains pairs of Dhivehi sentences: original correct sentences and their synthetic error-containing counterparts. The dataset contains approximately 7.2M sentence pairs (split into train/validation/test), where each pair consists of a correct Dhivehi sentence and the same sentence with synthetic errors. Errors are synthetically generated using rule-based transformations including diacritic errors, character mistakes, tense marker errors, case marking mistakes, word order issues, agreement errors, and suffix errors.