러시아어 형태소 분석기 MyStem, Pymorphy, TreeTagger 성능 비교연구 －신조어와 동음이의어를 중심으로－

초록

This study compares the practical performance of three Russian morphological analyzers-MyStem, Pymorphy, and TreeTagger—through experiments focusing on the analysis of neologisms and homonymous words. The results of the neologism analysis show that MyStem generates lemmas for almost all neologisms and tends to place them within existing Russian inflectional paradigms, presenting a wide range of possible grammatical features simultaneously. Pymorphy, which combines rule-based processing with probabilistic models, produces relatively consistent outputs by selecting a single most probable analysis; however, it occasionally misclassifies neologisms as proper nouns. TreeTagger provides stable part-of-speech and basic grammatical information, but often fails to generate lemmas for out-of-vocabulary words. In the homonym analysis, both MyStem and Pymorphy exhibit repeated misanalyses in sentences containing homonymous forms due to their lack of context-based disambiguation, whereas TreeTagger achieves the highest accuracy at the part-of-speech level by exploiting contextual information. Nevertheless, TreeTagger also shows errors in finer-grained grammatical categories and in the determination of verbal aspect. These findings indicate that the three morphological analyzers exhibit different strengths and limitations in handling neologisms and resolving ambiguity, and that the choice of an appropriate analyzer should depend on the research goals and the characteristics of the data.

키워드