A Rule System for Chinese Time Entity Recognition by Comprehensive Linguistic Study

International Joint Conference on Natural Language Processing, pages 795–801, Nagoya, Japan, 14-18 October 2013. A Rule System for Chinese Time Entity Recognition by Comprehensive Linguistic Study Hongzhi Xu The Department of CBS The Hong Kong Polytechnic University hongz.xu@gmail.com Chu-Ren Huang Faculty of Humanities The Hong Kong Polytechnic University churenhuang@gmail.com Abstract Chinese time entity is quite complex. In this paper, we give a comprehensive lin- guistic study on it. Based on the analy- sis, we present a rule system which on- ly considers the inner structure of Chinese time entities for the recognition. Exper- iments on Sinica and TempEval-2 corpus show that the rule system performs much better than the CRFs model. When using the rules as features within a CRFs mod- el, the performance could be further im- proved. 1 Introduction In SemEval-2010 competition, there is a sub task for temporal entity identification, which includes a Chinese corpus. The final goal of the task is to associate a temporal expression to a certain even- t. It is very important to extract all the elements for events in that it will be useful for event track- ing. By identifying the time information of events will enable us to make inference on the temporal relation of different events. In this paper, we will make a comprehensive s- tudy on Chinese time entities from a linguistic per- spective and then present a rule system for recog- nizing them. Chinese temporal entity is very com- plex due to the flexible grammar of Chinese and the existence of many different time systems, such as Gregorian system, the Chinese lunar system, the Chinese tian-gan & di-zhi (GZ) time system. Based on our linguistic analysis, we formal- ize a set of temporal elements that are the blocks used to construct time entities, such as century, year, month, day, hour etc. We then build a rule system that actually describe the topol- ogy of the temporal elements. For example, year follows century; month follows year. So, the model of our system is a directed graph, while a valid temporal expression is a path from one cer- tain node to another node. The longer the path is, the more confident the recognition will be. CTEMP (Wu et al., 2005) also used linguis- tic rules for Chinese temporal entity recognition. However, the focus of this work differs from them in that we aims to identify Chinese time entities which could be described with a limited set of rules and can be easily translated into a structured format, such as TIMEX3(Pustejovsky et al., 2010) standard. For this part, the set of rules in this work are more comprehensive than (Wu et al., 2005). However, we don’t include events that are used as time entities, since events intrinsically are not time entities. According to the Generative Lexi- con Theory (Pustejovsky, 1995), this is a case of type coercion. In Section 2, we will give a linguistic study on Chinese time entity expressions. In Section 3, we will construct a rule system which …

Content truncated to 500 words