KurdishMCQ: A multiple-choice question dataset for the Kurdish language (Sorani dialect)

Published: 14 September 2026| Version 2 | DOI: 10.17632/z8z28jhszp.2
Contributors:
,

Description

KurdishMCQ is, to the authors' knowledge, the first large-scale, publicly available collection of multiple-choice questions in the Kurdish language (Sorani dialect), supporting the evaluation of large language models in low-resource settings. Version 2 contains 17,221 questions digitized from Kurdish school textbooks, previous-year ministerial examinations, and public driving-test questions, all in Sorani (Perso-Arabic script), created through an OCR pipeline and verified by native Kurdish speakers. Each record includes a stable identifier, the question text, up to four answer options, the correct answer key, subject, grade, educational group, and provenance (source document and, for textbook items, the printed page number). Subjects span STEM, social sciences, humanities, and Kurdish language; most questions come from grade 12. See the included README for the full data dictionary, composition, usage guidance, and the V2 change log; the quality-control reports document every correction made in this version, and an externally validated random sample (198/200 fully correct; 95% CI 96.4–99.7%) characterizes accuracy. Written permission for digitizing the textbook exercises and ministerial examination questions and releasing them under CC BY 4.0 was granted by the General Directorate of Education of Sulaymaniyah, Ministry of Education, Kurdistan Regional Government (letter No. 16914, September 9, 2026). Processing and audit scripts: https://github.com/khalidchawtany/KurdishMCQ

Files

Institutions

Categories

Natural Language Processing, Large Language Model

Licence