통계적 방법에 의한 한국어 텍스트 유형 및 문체 분석

Abstract

This study aims to analyze Korean texts of various genres by means of computational and statistical methods. We collected 334 text samples of 36 genres of Korean texts. Each text sample is a computerized one and is composed of about 1,000 words, the total size of the corpus being about 370,000 words. Adopting Biber`s (1988) multivariate and multidimensional methods of text analysis, we counted the occurrences of 82 linguistics features of each text samples and performed factor analysis, cluster analysis, and canonical discriminant analysis. The findings suggest that there are 6 dimensions of Korean texts: Informal Interaction vs Elaboration, On-line/Situated Production vs. Informative Content, Narrative vs. Abstract Concerns, Formal Statement of Opinions, Public Report, and Public Mention of Modern Things. This is the result of factor analysis. By cluster analysis, we have found that there are 8 text types which are solely based on linguistic characteristics and which are different from conventional genres. The canonical discriminant analysis, which considers conventional genres seriously, does not show a clear picture but it conforms to the factor analysis and cluster analysis which show that real conversations (or the transcription of conversations) are really different from modeled, idealized conversations founded in scripts of dramas, plays, cinemas, novels, and comics.