This Korean text corpus is built from professionally published books, cleaned for training use and spanning the humanities, social sciences, natural sciences, and practical subjects.
Potential Use Cases
- Korean Pre-Training: Editorially produced book text supplies consistent orthography, punctuation, and paragraph structure at a quality level that crawled Korean text rarely reaches.
- Domain Balancing of a Training Mix: Coverage across four subject areas lets a Korean pre-training corpus be weighted by field on purpose, instead of inheriting whatever distribution a crawl happened to collect.
- Long-Context and Document-Level Training: Book-length material provides passages long enough to train and evaluate long-context comprehension and document-level summarization in Korean.