Joint Inference for Bilingual Semantic Role Labeling

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 304–314, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Joint Inference for Bilingual Semantic Role Labeling Tao Zhuang and Chengqing Zong National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences {tzhuang, cqzong}@nlpr.ia.ac.cn Abstract We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce ar- gument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilin- gual SRL result that is not only reasonable on each side of bitext, but also has more consis- tent argument structures between two sides. To evaluate the consistency between two argu- ment structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel Prop- Bank data. Using our joint inference model, F1 scores of SRL results on Chinese and En- glish text achieve 79.53% and 77.87% respec- tively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively. 1 Introduction In recent years, there has been an increasing inter- est in SRL on several languages. However, little research has been done on how to effectively per- form SRL on bitext, which has important applica- tions including machine translation (Wu and Fung, 2009). A conventional way to perform SRL on bi- text is performing SRL on each side of bitext sep- arately, as has been done by Fung et al. (2007) on Chinese-English bitext. However, it is very difficult to obtain good SRL results on both sides of bitext in this way. The reason is that even the state-of- the-art SRL systems do not have very high accuracy on both English text (M`arquez et al., 2008; Pradhan et al., 2008; Punyakanok et al., 2008; Toutanova et al., 2008), and Chinese text (Che et al., 2008; Xue, 2008; Li et al., 2009; Sun et al., 2009). On the other hand, the semantic equivalence be- tween two sides of bitext means that they should have consistent predicate-argument structures. This bilingual argument structure consistency can guide us to find better SRL results. For example, in Fig- ure 1(a), the argument structure consistency can guide us to choose a correct SRL result on Chinese side. Consistency between two argument structures is reflected by sound argument alignments between them, as shown in Figure 1(b). Previous research has shown that bilingual constraints can be very help- ful for parsing (Burkett and Klein, 2008; Huang et al., 2008). In this paper, we show that the bilingual argument structure consistency can be leveraged to substantially improve SRL results on both sides of bitext. Formally, we present a joint inference model to preform bilingual SRL. Using automatic word align- ment on bitext, we first identify a pair of predicates that …

Content truncated to 500 words