Automatic Inference of the Tense of Chinese Events Using Implicit Linguistic Information

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1902–1911, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Automatic Inference of the Tense of Chinese Events Using Implicit Linguistic Information Yuchen Zhang Brandeis University 415 South Street Waltham, MA yuchenz@brandeis.edu Nianwen Xue Brandeis University 415 South Street Waltham, MA xuen@brandeis.edu Abstract We address the problem of automatically inferring the tense of events in Chinese text. We use a new corpus annotated with Chinese semantic tense information and other implicit Chinese linguistic informa- tion using a “distant annotation” method. We propose three improvements over a rel- atively strong baseline method – a statisti- cal learning method with extensive feature engineering. First, we add two sources of implicit linguistic information as fea- tures – eventuality type and modality of an event, which are also inferred automat- ically. Second, we perform joint learning on semantic tense, eventuality type, and modality of an event. Third, we train arti- ficial neural network models for this prob- lem and compare its performance with feature-based approaches. Experimental results show considerable improvements on Chinese tense inference. Our best per- formance reaches 68.6% in accuracy, out- performing a strong baseline method. 1 Introduction As a language with no grammatical tense, Chinese does not encode the temporal location of an event directly in a verb, while in English, the grammati- cal tense of a verb is a strong indicator of the tem- poral location of an event. In this paper we ad- dress the problem of inferring the semantic tense, or the temporal location of an event (e.g., present, past, future) in Chinese text. The semantic tense is defined relative to the utterance time or document creation time, and it does not always agree with the grammatical tense in languages like English where there is grammatical tense. Inferring se- mantic tense potentially benefits natural language processing tasks such as Machine Translation and Information Extraction (Xue, 2008; Reichart and Rappoport, 2010; Ye et al., 2006; Ye, 2007; Liu et al., 2011), but previous work has shown that auto- matic inference of the semantic tense of events in Chinese is a very challenging task (Xue, 2008; Ye et al., 2006; Liu et al., 2011). There are at least two reasons why this is a dif- ficult problem. First, since Chinese does not have grammatical tense which could serve as an impor- tant clue when annotating the semantic tense of an event, generating consistent annotation for Chi- nese semantic tense has proved to be a challenge. Xue and Zhang (2014) use a “distant annotation” method to address this problem. They take advan- tage of an English-Chinese parallel corpus with manual word alignments (Li et al., 2012) , and per- form annotation on the English side, which pro- vides more explicit information such as grammati- cal tense that helps annotators decide the appropri- ate semantic tense. The annotations are then pro- jected to the Chinese side via the word alignments. They show consistent annotation agreements on semantic …

Content truncated to 500 words