Korean TimeBank Including Relative Temporal Information

Korean TimeBank Including Relative Temporal Information Chae-Gyun Lim†, Young-Seob Jeong§, Ho-Jin Choi† †School of Computing, KAIST 291 Daehak-ro, Yuseong-gu, Daejeon 34141, South Korea {rayote, hojinc}@kaist.ac.kr §Department of BigData Engineering, Soonchunhyang University 22 Soonchunhyang-ro, Asan-si, Chungcheongnam-do 31538, South Korea bytecell@sch.ac.kr Abstract Since most documents have temporal information that can be a basis for understanding the context, the importance of temporal information extraction researches is steadily growing. Although various attempts have been made to extract temporal information from researchers internationally, it is difficult to apply them to other languages because they are usually targeted at specific languages such as English. Several annotation languages and datasets had been proposed for the studies of temporal information extraction on Korean documents, however, the representation of relative temporal information is not enough to maintain it explicitly. In this paper, we propose a concept of relative temporal information and supplement a Korean annotation language to represent new relative expressions, and extend an annotated dataset, Korean TimeBank, through the revised language. We expect that it is possible to utilize potential features from the Korean corpus by clearly annotating relative temporal relationships and to use well-refined Korean TimeBank in future studies. Keywords: Relative temporal information, Korean TimeML, Relative temporal relationship, Korean TimeBank 1. Introduction Temporal information extraction is one of the important re- search fields in natural language processing, and it is neces- sary to understand the temporal context in real-world appli- cations such like question answering or conversation sys- tem providing a good quality of service. As an interna- tional activity related to temporal information extraction re- search, there is a shared task called TempEval which is part of SemEval (UzZaman et al., 2012). According to the de- scription of TempEval, the temporal information extraction consists of three separate processes as follows—timex3, event, and tlink extraction. TempEval has provided an- notated datasets for performance comparison by adopting TimeML (Pustejovsky et al., 2003), which is one of well- known markup languages for annotating temporal informa- tion. However, since the dataset is basically constructed un- derlying English documents, it is not applicable to the stud- ies of temporal information extraction in other languages. Preceding the construction of a corpus consisting of the documents in a particular language is very important be- cause the linguistic characteristics inherent in the language have a significant impact to discover temporal relationships (Jeong, 2016). For extracting temporal information in Korean, there are several previous researches. The Korean TimeML (Im et al., 2009), which is adopt morpheme-level stand-off anno- tation scheme and addressed some language-specific issues on Korean, was proposed. In 2015, the revised version of Korean TimeML was proposed to overcome several limi- tations such as applying the lunar calendar and character- level annotation scheme (Jeong et al., 2015). Moreover, they also published a corpus, namely Korean TimeBank, including a bunch of Korean documents annotated by the revised Korean TimeML. In this paper, we propose a concept to find relative temporal information from Korean documents and complement the previous version of Korean TimeML to be …

Content truncated to 500 words