접사 정보를 이용한 형태소 분석 연구

Abstract

The aim of this paper is to propose the morphological analysis system using the affix information. Based on rule-based methodology, the morphological analysis system is constructed. Many of the analysis errors are caused by unregistered words. If target words('euseul') is composed of unregistered words, too many ending analyses are easy to make an error. Especially, the words with the affix which is one syllable word is difficult to analyze correctly. Such errors are able to be corrected by using affix information. The affix property information and affix combination rule prevent the morphological analysis system from making erroneous results. The affix combination rule is a kind of the feature checking rule. If the ending has [v] and the stem has [v], the analysis is correct, and vice versa. But some words do not follow the rules. As the situation is very limited, we can make simple filtering rules. The Affix property information is related to the combined place. Some affixes are always combined with adjacent place to the stem. Some affixes have some restriction to the combination to the adjacent place. And some affixes are placed in the rightmost place of the word. By these property, we can assign the special feature to the affixes. These features can be used to analyze the words and the success rate of the morphological analysis will pick up considerably. The affix rules which are made in this paper are very compact, the application to the real system is not difficult. Especially, as these rules are composed of the independent modules, the expansion and amendment of the rules are easy, which is the merit of this system.