Chinese Semantic Role Labeling with Shallow Parsing

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1475–1483, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Chinese Semantic Role Labeling with Shallow Parsing Weiwei Sun and Zhifang Sui and Meng Wang and Xin Wang Institute of Computational Linguistics Peking University Key Laboratory of Computational Linguistics Ministry of Education, China weiwsun@gmail.com;{szf,wm}@pku.edu.cn;xinwang.cpku@gmail.com; Abstract Most existing systems for Chinese Seman- tic Role Labeling (SRL) make use of full syntactic parses. In this paper, we evalu- ate SRL methods that take partial parses as inputs. We first extend the study on Chi- nese shallow parsing presented in (Chen et al., 2006) by raising a set of addi- tional features. On the basis of our shal- low parser, we implement SRL systems which cast SRL as the classification of syntactic chunks with IOB2 representation for semantic roles (i.e. semantic chunks). Two labeling strategies are presented: 1) directly tagging semantic chunks in one- stage, and 2) identifying argument bound- aries as a chunking task and labeling their semantic types as a classification task. For both methods, we present encouraging re- sults, achieving significant improvements over the best reported SRL performance in the literature. Additionally, we put forward a rule-based algorithm to auto- matically acquire Chinese verb formation, which is empirically shown to enhance SRL. 1 Introduction In the last few years, there has been an increas- ing interest in Semantic Role Labeling (SRL) on several languages, which consists of recognizing arguments involved by predicates of a given sen- tence and labeling their semantic types. Nearly all previous Chinese SRL research took full syn- tactic parsing as a necessary pre-processing step, such as (Sun and Jurafsky, 2004; Xue, 2008; Ding and Chang, 2008). Many features are extracted to encode the complex syntactic information. In En- glish SRL research, there have been some attempts at relaxing the necessity of using full syntactic parses; better understanding of SRL with shallow parsing is achieved by CoNLL-2004 shared task (Carreras and M`arquez, 2004). However, it is still unknown how these methods perform on other lan- guages, such as Chinese. To date, the best SRL performance reported on the Chinese Proposition Bank (CPB) corresponds to a F-measure is 92.0, when using the hand- crafted parse trees from Chinese Penn Treebank (CTB). This performance drops to 71.9 when a real parser is used instead1 (Xue, 2008). Com- paratively, the best English SRL results reported drops from 91.2 (Pradhan et al., 2008) to 80.56 (Surdeanu et al., 2007). These results suggest that as still in its infancy stage, Chinese full parsing acts as a central bottleneck that severely limits our ability to solve Chinese SRL. On the contrary, Chi- nese shallow parsing has gained a promising re- sult (Chen et al., 2006); hence it is an alternative choice for Chinese SRL. This paper addresses the Chinese SRL problem on the basis of shallow syntactic information at the level of phrase chunks. We first extend the study on Chinese chunking presented in (Chen et al., 2006) by raising a …

Content truncated to 500 words