Open Korean Corpora: A Practical Report

Korean is often labeled a low-resource language in computational linguistics, but a new survey argues the label stems as much from poor curation and advertising of existing datasets as from any genuine scarcity [1]. The paper, authored by Won Ik Cho, Sangwhan Moon, and Youngsook Song, was presented at the 2020 Workshop for NLP Open Source Software and updated through June 2026 [1][3]. It curates a list of Korean corpora, first describing institution-level resource development and then cataloguing open datasets for tasks such as parsing, tagging, and other text-mining applications [1][5]. The authors note that many high-quality Korean datasets are produced by domestic institutions but remain difficult for researchers outside the country to access. “Researchers abroad can indeed access the data, but they may face difficulty filling out and submitting the particular application form, instead of the barrier-free downloading system,” the paper states [5]. The survey introduces three checklists to evaluate each corpus: documentation, usage terms, and redistribution rights [5]. It highlights the KAIST Morpho-Syntactically Annotated Corpus, which applies morphological analysis to a freely available raw corpus of about 70 million words spanning novels, non-literature texts, and articles [5]. The paper also tracks the evolution of its own preprint: the first submission on arXiv in December 2020 was 43 KB, a May 2023 revision grew to 61 KB, and the June 2026 version expanded to 578 KB, reflecting the addition of new datasets over time [1]. Text mining, the broader field in which such corpora are used, involves structuring unstructured text to extract patterns, categorize documents, or model entity relationships [7]. The availability of well-documented, openly licensed corpora directly affects the quality of these downstream tasks. The authors propose that open-source dataset construction for less-resourced languages should follow a model where documentation is fine-grained, usage terms are clear, and the outcome is internationally available [5]. The paper frames this as a practical path to promoting research in languages that are currently underserved by the NLP community [1][3].