한국어 체언말 저해음 변이를 위한 복수기저형 확률문법

초록

This paper aims to propose a probabilistic grammar with multiple underlying representations (UR’s) for Korean noun-final obstruent variation and to analyze the influence of UR choice on this variation. The grammar is based on Pater et al.’s (2012) multiple UR Maximum Entropy model, where UR constraints are proposed to require a particular UR corresponding to the meaning/concept of a given morpheme or word. It consists of weighted constraints learned from the training data which are a result of the picture-presented production experiment conducted for this variation. This grammar produces the probabilitic distribution of variation for Korean obstruent-final nouns. The probabilistic distribution tells us that this variation consists of three different types of variation. One is a (traditional or etymological) UR-preferred type of variation (ch-final and ph-final nouns), another is a substitution-preferred one (kh-final nouns) and the other is the variation which a phonological process strongly influences (th-final nouns). This phenomenon shows that the variation is difficult to explain without considering the UR factor. Further, in this paper, it is shown that the UR choice is determined by the interaction between UR constraints, faithfulness constraints and markedness constraints.

키워드