MultiGEC is a dataset for Grammatical Error Correction containing parallel data for 12 languages and 17 subcorpora. Each subcorpus contains two or more parallel versions of the same texts (typically, full learner essays), where one version (orig) is the one that the author originally wrote, and the others (ref1, ref2, ...) are corrected versions of the same text. Languages included: Czech, English, Estonian, German, Greek, Icelandic, Italian, Latvian, Russian, Slovene, Swedish and Ukrainian (English and Russian are available on request). Texts come from different original corpora, but are reformatted to a unified format.
Dataset description
MultiGEC is a dataset for Multilingual Grammatical Error Correction in 12 European languages (Czech, English, Estonian, German, Greek, Icelandic, Italian, Latvian, Russian, Slovene, Swedish and Ukrainian) compiled by the CompSLA working group and over 20 external data providers in the context of MultiGEC-2025, the first text-level GEC shared task.
The MultiGEC dataset is divided into 17 subcorpora covering different languages, domains and correction styles, summarized below. More detailed information about each subcorpus is available as machine-readable metadata, whose format is described .