Korean Morphological Analysis with Tied Sequence-to-Sequence Multi-Task Model

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1436–1441, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1436 Korean Morphological Analysis with Tied Sequence-to-Sequence Multi-Task Model Hyun-Je Song Dept. of Information Technology Chonbuk National University Jeonju, 54896, Korea hyunje.song@jbnu.ac.kr Seong-Bae Park Dept. of Computer Science and Engineering Kyung Hee University Yongin, 17104, Korea sbpark71@khu.ac.kr Abstract Korean morphological analysis has been con- sidered as a sequence of morpheme pro- cessing and POS tagging. Thus, a pipeline model of the tasks has been adopted widely by previous studies. However, the model has a problem that it cannot utilize interactions among the tasks. This paper formulates Ko- rean morphological analysis as a combination of the tasks and presents a tied sequence-to- sequence multi-task model for training the two tasks simultaneously without any explicit reg- ularization. The experiments prove the pro- posed model achieves the state-of-the-art per- formance. 1 Introduction Korean is an agglutinative language (Song, 2006). Thus, it is a fundamental step for understanding a sentence to analyze the grammatical structure of eojeols, where an eojeol is a linguistic unit seg- mented by a white space. An eojeol is composed of one or more morphemes. As a result, one eojeol can be analyzed into several morpheme combina- tions depending on a context, which yields dif- ferent part-of-speech (POS) tags of the morpheme combinations. In addition, some morphemes have a different surface form from their base form when they are derived from an eojeol. Therefore, the goal of Korean morphological analyzer is not only to decompose and recover morphemes from eoje- ols precisely (morpheme processing), but also to assign POS tags to the decomposed and/or recov- ered morphemes accurately according to a context (POS tagging). Traditional approaches to Korean morphologi- cal analysis have adopted a pipeline model of mor- pheme processing and POS tagging (Lee and Rim, 2009; Na, 2015; Choi et al., 2016; Matteson et al., 2018; Song and Park, 2018). That is, they first �� � � �� � NNG JKO NP JX VV ETM � � � � � � � NNG JKO VV EP EF Eojoel sequence POS tag sequence Morpheme sequence (I) (in the sky) (flying) (a bird) (saw) ��� �� �� �� Figure 1: Korean morphological analysis of a sentence “나는 하늘을 나는 새를 봤다” of which meaning is “I saw a bird flying in the sky”. Correct morpheme analysis helps predicting POS tags (blue dotted arrows) while POS tagging affects morpheme analysis (red ar- rows). Orange morphemes in the morpheme sequence are recovered morphemes. Best viewed in color. decompose and recover morphemes from eoje- ols or assign so-called POSMORPH tags (Heigold et al., 2016), and then an actual POS tag sequence is determined or resolved from the POSMORPH tags using a sequential labeling algorithm. How- ever, this pipeline model suffers from two kinds of weaknesses. One weakness is that the errors from morpheme processing are apt to be propagated …

Content truncated to 500 words