말뭉치언어학의 관점에서 본 <모두의 말뭉치> 구축 현황과 쟁점

초록

This study examined the status of the National Institute of Korean Language’s Everyone’s Corpus and explored key issues in and future directions for building text data in the era of generative artificial intelligence (“AI”). To this end, six analytical criteria were established: purpose of construction, text genre, level and type of processing, multilingual parallelism, and data type. As of March 31, 2025, Everyone’s Corpus includes 82 corpora constituted by 2.7 billion words and various web-based corpus tools. While the corpus is meaningful for its large size and can be used to train and evaluate AI models, over 85% of its content consists of written language, indicating a clear imbalance in text genre representation. Thus, this paper discusses the relevance of traditional balanced corpora such as COCA as benchmarks for generative language models and proposes the construction of inclusive corpora that reflect speaker subjectivity and guarantee linguistic rights for visually and hearing-impaired people.

키워드