Shallow Semantic Parsing of Chinese

Shallow Semantic Parsing of Chinese Honglin Sun1 Center for Spoken Language Research University of Colorado at Boulder Daniel Jurafsky2 Center for Spoken Language Research University of Colorado at Boulder Abstract In this paper we address the question of assigning semantic roles to sentences in Chinese. We show that good semantic parsing results for Chinese can be achieved with a small 1100-sentence training set. In order to extract features from Chinese, we describe porting the Collins parser to Chinese, resulting in the best performance currently reported on Chinese syntactic parsing; we include our head- rules in the appendix. Finally, we compare English and Chinese semantic-parsing performance. While slight differences in argument labeling make a perfect comparison impossible, our results nonetheless suggest significantly better performance for Chinese. We show that much of this difference is due to grammatical differences between English and Chinese, such as the prevalence of passive in English, and the strict word order constraints on adjuncts in Chinese. 1 Introduction Thematic roles (AGENT, THEME, LOCATION, etc) provide a natural level of shallow semantic representation for a sentence. A number of algorithms have been proposed for automatically assigning such shallow semantic structure to English sentences. But little is understood about how these algorithms may perform in other languages, and in general the role of language-specific idiosyncracies in the extraction of semantic content and how to train these algorithms when large hand-labeled training sets are not available. In this paper we address the question of assigning semantic roles to sentences in Chinese. Our work is 1 Currently at Department of Computer Science, Queens College, City University of New York. Email: sunh@qc.edu. 2 Currently at Department of Linguistics, Stanford University. Email: jurafsky@stanford.edu. based on the SVM-based algorithm proposed for English by Pradhan et al (2003). We first describe our creation of a small 1100-sentence Chinese corpus labeled according to principles from the English and (in-progress) Chinese PropBanks. We then introduce the features used by our SVM classifier, and show their performance on semantic parsing for both seen and unseen verbs, given hand-corrected (Chinese TreeBank) syntactic parses. We then describe our port of the Collins (1999) parser to Chinese. Finally, we apply our SVM semantic parser to a matching English corpus, and discuss the differences between English and Chinese that lead to significantly better performance on Chinese. 2 Semantic Annotation and the Corpus Work on semantic parsing in English has generally related on the PropBank, a portion of the Penn TreeBank in which the arguments of each verb are annotated with semantic roles. Although a project to produce a Chinese PropBank is underway (Xue and Palmer 2003), this data is not expected to be available for another year. For these experiments, we therefore hand-labeled a small corpus following the Penn Chinese Propbank labeling guidelines (Xue, 2002). In this section, we first describe the semantic roles we used in the annotation and then introduce the data for our experiments. 2.1 Semantic roles Semantic roles in the English (Kingsbury et al 2002) and Chinese (Xue …

Content truncated to 500 words