Multi-Predicate Semantic Role Labeling

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 363–373, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Multi-Predicate Semantic Role Labeling Haitong Yang and Chengqing Zong National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China {htyang, cqzong}@nlpr.ia.ac.cn Abstract The current approaches to Semantic Role Labeling (SRL) usually perform role clas- sification for each predicate separately and the interaction among individual predi- cate’s role labeling is ignored if there is more than one predicate in a sentence. In this paper, we prove that different predi- cates in a sentence could help each other during SRL. In multi-predicate role label- ing, there are mainly two key points: argu- ment identification and role labeling of the arguments shared by multiple predicates. To address these issues, in the stage of argument identification, we propose nov- el predicate-related features which help re- move many argument identification errors; in the stage of argument classification, we adopt a discriminative reranking approach to perform role classification of the shared arguments, in which a large set of glob- al features are proposed. We conducted experiments on two standard benchmarks: Chinese PropBank and English PropBank. The experimental results show that our approach can significantly improve SRL performance, especially in Chinese Prop- Bank. 1 Introduction Semantic Role Labeling (SRL) is a kind of shal- low semantic parsing task and its goal is to rec- ognize some related phrases and assign a joint structure (WHO did WHAT to WHOM, WHEN, WHERE, WHY, HOW) to each predicate of a sen- tence (Gildea and Jurafsky, 2002). Because of the ability of encoding semantic information, SR- L has been applied in many tasks of NLP, such as question and answering (Narayanan and Haraba- gir, 2004), information extraction (Surdeanu et The justices will be forced to reconsider the questions. [ A1 ] [ Pred ] [ A0 ] [ Pred ] [ A1 ] Figure 1: A sentence from English PropBank, with an argument shared by multiple predicates al., 2003; Christensen et al., 2005), and machine translation (Wu and Fung, 2009; Liu and Gildea, 2010; Xiong et al., 2012; Zhai et al., 2012). Currently, an SRL system works as follows: first identify argument candidates and then per- form classification for each argument candidate. However, this process only focuses on one inde- pendent predicate without considering the internal relations of multiple predicates in a sentence. Ac- cording to our statistics, more than 80% sentences in Propbank carry multiple predicates. One exam- ple is shown in Figure 1, in which there are two predicates ‘Force’ and ‘Reconsider’. Moreover, the constituent ‘the justices’ is shared by the two predicates and is labeled as A1 for ‘Force’ but as A0 for ‘Reconsider’. We call this phenomenon of the shared arguments Role Transition . Intuitive- ly, all predicates in a sentence are closely related to each other and the internal relations between them would be helpful for SRL. This paper has made deep investigation on multi-predicate semantic …

Content truncated to 500 words