Machine-learning and corpus-based analyses of Korean nouns from subjects and objects: Stimuli development of verb treatment for neurogenic communication disorders [신경언어장애군의 동사 중재 자극 개발을 위한 기계학습 및 빅데이터 기반 한국어의 주격 및 목적격 명사 분석]

Abstract

Objectives: Subjects and objects associated with treatment verbs are common stimuli in verb treatment using the argument structure for neurogenic patients. This study investigated Korean nouns from subjects and objects associated with the target verbs in the corpus to suggest the characteristics of subjects and objects for developing Korean-specific verb treatment stimuli. In addition, we shared raw data through cloud so that anyone can use the data for clinical or academic purposes. Methods: We used Korean textbook corpus to investigate the differences between subjects and objects in terms of frequency, number of type, and machine-learning based semantic distance to the target verbs. We also examined how machine-learning based semantic distance is correlated with behavioral rating semantic distance. Results: Subjects significantly showed less frequency, less number of types, and farther machine-learning based distance to the target verbs than objects did. Machine-learning based semantic distance was strongly correlated with behavioral rating. Conclusion: The results demonstrated strong evidence of the subject ellipsis phenomenon in Korean. The weak semantic relation, as proven by the machine-learning based semantic distance, indicated that subjects as verb stimuli for Korean-speaking neurogenic patients need to be reconsidered. We confirmed the possibility of machine-learning based semantic distance as a criterion in selecting the treatment stimuli, but more detailed verification is required for future studies. © 2019 Korean Academy of Speech-Language Pathology and Audiology.