Extensive Use of Morpheme Features in Korean Dependency Parsing

Hwijeen Ahn, Minyoung Seo, Chanmin Park, Juae Kim, Jungyun Seo

Abstract: Dependency parsing is considered to be one of the most important tasks in natural language understanding. With the advance of deep learning based models, dependency parsing performances were improved. However, little attention was given to features specific to Korean. Morphemes play especially important role in Korean linguistics, and incorporating morpheme level information into the model can result in performance improvement. In this paper, we propose a way to use multiple morpheme features with their orders preserved in Korean dependency parsing. In our experiment, our proposed method gave a performance improvement in deep biaffine network and stack pointer network. We also report the state-of - the-art result of UAS 92.17 and LAS 90.07 on Sejong dependency parsing data with stack pointer network augmented with our proposed method.