An Empirical Study of Zero Anaphora Resolution in Chinese Based on Centering Model

1 An Empirical Study of Zero Anaphora Resolution in Chinese Based on Centering Model Ching-Long Yeh and Yi-Jun Chen Department of Computer Science and Engineering Tatung University 40 Chungshan N. Rd. 3rd. Section Taipei 104 R.O.C. chingyeh@cse.ttu.edu.tw d8806005@cseserv.cse.ttu.edu.tw Abstract In this paper, we describe the creation of Chinese zero anaphora resolution rules by performing experiments. The rules were constructed based on the centering model. In the experiments, we selected several texts as testing examples. We compared the referents of zero anaphors in the testing texts identified by hand with the ones resolved by using an algorithm employing a resolution rule. Three rules were used to carry out the experiment. The results show that the rule considering grammatical role criteria and domain knowledge obtained the best result: 85% of zero anaphors in the test texts were correctly resolved. We investigate problems of miss-resolution of zero anaphors in the test text and propose solution to deal with them. 1. Introduction In Chinese text, anaphors are frequently eliminated, termed zero anaphor (ZA) hereafter, due to their prominence in discourse [LT81]. For example in (1), the topic of the utterance (1a) is 電子股 ‘ Electronics stocks,’ which is eliminated in the second utterance and the topic of utterance (1c), 證券股 ‘Securities stocks,‘ is eliminated in the utterance (1d). (1) a. 電子股 i 受 美國高科技股 重挫 影響， Electronics stocks were affected by high-tech stocks fallen heavily in America. b. φ i 今日 持續 下跌； (Electronics stocks) continued falling down today. c. 證券股 j 也 相對回應， Securities stocks also had respondence. d. φ j 盤 中 陸續 下殺 至 跌停。 (Securities stocks) fell by close one after another on the market. A simple rule, Rule 1, can be formulated by observing the phenomenon of topic chain in Chinese text. This rule can be used to correctly resolve the referent of the ZA in 2 (1a), for example. Rule 1: If a ZA occurs in the topic position of utterance i, then its antecedent is the topic of utterance i-1. In general, zero anaphors in Chinese can occur in any grammatical slot with an antecedent that may occur in any grammatical slot, regardless of their distance [LT79]. Thus Rule 1 is obviously insufficient to account for the resolution of ZAs. Within the theories of discourse, Centering is a computational model, which has been developed as a methodology for the explanation of the local coherence and its relationship to attentional state at the local level and focuses on pronominal and nominal anaphora [GJW83, GJW95]. It is formalized as a system of constraints and rules, which can, as part of a computational discourse model, act to control inference [JW81]. In the centering model, each utterance in a discourse segment has two structures associated with it, called Forward-Looking and Backward-Looking centers, which correspond approximately to Sidner's potential foci and discourse focus [Sid79]. Forward-Looking Centers, Cf, is a set of discourse entities in an utterance, and Backward-Looking Center, Cb, is a special member of this set, which is the discourse entity …

Content truncated to 500 words