Korean Time Information Analysis of Hierarchical Annotation Rules from Natural Language Text

Abstract

Documents contain temporal information that can be used for various applications such as recommendation systems. In order to find the information, analysis of time expressions is important from the documents written by natural language texts. As inherent linguistic characteristics of the Korean language, it is hard to capture every time expression from a given text without considering various patterns of temporal information. In this paper, we propose a hierarchical patterns to annotate time information that can reflect the characteristics in Korean language texts. We define 12 patterns of Korean time expression and 5 types of pattern groups which are hierarchically referred for the given text including digits, number expressions, particular expressions (e.g., season, day of week, etc.), event patterns, and Lunar holidays. Also, we study on their effects by analyzing the frequencies and coverages in a human-annotated corpus.