[DOI: 10.17959/sppm.2025.31.3.371]

[본문 파일]

음소배열제약기반의 한국어 어종 분류: 통계적 문법 모델 예측과 화자의 판단 비교

박나영 (서울대학교)

Abstract 
Phonotactic constraints in Korean lexical classification: A comparison of statistical grammar model predictions and speaker judgments. This study tests the hypothesis that phonotactic constraints distinguish Korean lexical classes — native Korean, Sino-Korean and loanwords — and that speakers employ these constraints in lexical class judgment. The hypothesis was tested in two ways. First, using multinomial logistic regression on 87,131 Korean nouns, we developed a statistical model that successfully classifies loanwords (F1=0.859) and Sino-Korean (F1=0.869), while native Korean showed limited phonotactic distinctiveness (F1=0.312). Second, a nonce-word judgment task with 64 speakers largely confirmed these patterns: speakers responded significantly to phonotactic cues for loanwords and Sino-Korean, while native Korean stimuli failed to elicit significant responses. These findings suggest that phonotactic constraints serve as effective cues for classifying loanwords and Sino-Korean, but not for native Korean.  

Keywords
phonotactic constraints, lexical classes, statistical phonotactic model, nonce-word judgment, Korean lexicon