The corpus is available in two JSON formats: ómarkað and markað. The ómarkað version contains parallel sentence pairs, with the original student sentence and the corrected version. This version does not include correction-level annotation or error codes. There are 8.723 different sentences in total. Out of these, 6.632 contain corrections and 2.091 contain no corrections.
The markað version contains the annotated correction data. It is structured with one correction per line. This means that a sentence with more than one correction is repeated once for each correction. In total, the markað file contains 13,292 lines: 11,628 lines with corrections and 1,664 lines without corrections. Counting only unique sentences, the corpus contains 5,872 different sentences: 4,208 with corrections and 1,664 without corrections.
A single correction may involve more than one error. When one correction depends on another, the two are linked and counted as one correction. For example, changing “, soleiðis” to “. Soleiðis” involves both changing the comma to a full stop and changing the following lowercase letter to uppercase, but this is counted as one correction in the data.
The resource also includes Faroese explanations of the error codes used in the data. These explanations can be provided upon request.
Because the corpus is based on student essays, access is restricted. The text is licensed and may not be redistributed or shared in any way. Access can be granted to partners of the Centre for Language Technology for internal research and training use only. To request access, users must contact the Centre for Language Technology and sign an agreement before the corpus can be shared.
Total words counted in corrected sentences: 155,553
Release: 26.09.2025
Contact: mtd@setur.fo
