Improving Chinese Semantic Role Classification with Hierarchical Feature Selection Strategy

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 324–333, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Improving Chinese Semantic Role Classification with Hierarchical Feature Selection Strategy Weiwei Ding Institute of Computational Linguistics Peking University Beijing, 100871, China weiwei.ding.pku@gmail.com Baobao Chang Institute of Computational Linguistics Peking University Beijing, 100871, China chbb@pku.edu.cn Abstract In recent years, with the development of Chi- nese semantically annotated corpus, such as Chinese Proposition Bank and Normalization Bank, the Chinese semantic role labeling (SRL) task has been boosted. Similar to Eng- lish, the Chinese SRL can be divided into two tasks: semantic role identification (SRI) and classification (SRC). Many features were in- troduced into these tasks and promising re- sults were achieved. In this paper, we mainly focus on the second task: SRC. After exploit- ing the linguistic discrepancy between num- bered arguments and ARGMs, we built a se- mantic role classifier based on a hierarchical feature selection strategy. Different from the previous SRC systems, we divided SRC into three sub tasks in sequence and trained models for each sub task. Under the hierarchical ar- chitecture, each argument should first be de- termined whether it is a numbered argument or an ARGM, and then be classified into fine- gained categories. Finally, we integrated the idea of exploiting argument interdependence into our system and further improved the per- formance. With the novel method, the classi- fication precision of our system is 94.68%, which outperforms the strong baseline signifi- cantly. It is also the state-of-the-art on Chi- nese SRC. 1 Introduction Semantic Role labeling (SRL) was first defined in Gildea and Jurafsky (2002). The purpose of SRL task is to identify and classify the semantic roles of each predicate in a sentence. The semantic roles are marked and each of them is assigned a tag which indicates the type of the semantic relation with the related predicate. Typical tags include Agent, Patient, Source, etc. and some adjuncts such as Temporal, Manner, Extent, etc. Since the arguments can provide useful semantic information, the SRL is crucial to many natural language proc- essing tasks, such as Question and Answering (Na- rayanan and Harabagiu 2004), Information Extrac- tion (Surdeanu et al. 2003), and Machine Transla- tion(Boas 2002). With the efforts of many re- searchers (Carreras and Màrquez 2004, 2005, Mo- schitti 2004, Pradhan et al 2005, Zhang et al 2007), different machine learning methods and linguistics resources are applied in this task, which has made SRL task progress fast. Compared to the research on English, the re- search on Chinese SRL is still in its infancy stage. Previous work on Chinese SRL mainly focused on how to transplant the machine learning methods which has been successful with English, such as Sun and Jurafsky (2004), Xue and Palmer (2005) and Xue (2008). Sun and Jurafsky (2004) did the preliminary work on Chinese SRL without any large semantically annotated corpus of Chinese. They just labeled the predicate-argument structures of ten specified verbs to a small collection of Chi- …

Content truncated to 500 words