Developing and validating lexical complexity indices for Korean

The relationship between lexical complexity and language production has been widely studied, with lexical complexity generally understood as a multidimensional construct. This study extends this line of research to Korean, a typologically distinct language in which lexical items cannot be straightforwardly defined based on assumptions derived from English. To address this issue, we propose a framework that operationalizes lexical items at both the eojeol (a whitespace-delimited unit) and morpheme levels and introduces three sub-dimensions of lexical complexity: diversity, sophistication, and compositionality. We validate this framework through two corpus-based experiments modeling different proficiency constructs in second language Korean writing. Across both experiments, the indices meaningfully predicted human-rated proficiency scores, explaining 18.2% of variance in vocabulary and grammatical use and 37.1% of the variance in holistic writing scores. All indices are available in the open-source Python package Korean Lexical Complexity Analyzer .