Semantic Role Tagging for Chinese at the Lexical Level

R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 804 – 814, 2005. © Springer-Verlag Berlin Heidelberg 2005 Semantic Role Tagging for Chinese at the Lexical Level Oi Yee Kwong and Benjamin K. Tsou Language Information Sciences Research Centre, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong {rlolivia, rlbtsou}@cityu.edu.hk Abstract. This paper reports on a study of semantic role tagging in Chinese, in the absence of a parser. We investigated the effect of using only lexical infor- mation in statistical training; and proposed to identify the relevant headwords in a sentence as a first step to partially locate the corresponding constituents to be labelled. Experiments were done on a textbook corpus and a news corpus, rep- resenting simple data and complex data respectively. Results suggested that in Chinese, simple lexical features are useful enough when constituent boundaries are known, while parse information might be more important for complicated sentences than simple ones. Several ways to improve the headword identifica- tion results were suggested, and we also plan to explore some class-based tech- niques for the task, with reference to existing semantic lexicons. 1 Introduction As the development of language resources progresses from POS-tagged corpora to syntactically annotated treebanks, the inclusion of semantic information such as predicate-argument relations is becoming indispensable. The expansion of the Penn Treebank into a Proposition Bank [11] is a typical move in this direction. Lexical resources also need to be enhanced with semantic information (e.g. [5]). In fact the ability to identify semantic role relations correctly is essential to many applications such as information extraction and machine translation; and making available re- sources with this kind of information would in turn facilitate the development of such applications. Large-scale production of annotated resources is often labour-intensive, and thus needs automatic labelling to streamline the work. The task can essentially be per- ceived as a two-phase process, namely to recognise the constituents bearing some semantic relationship to the target verb in a sentence, and then to label them with the corresponding semantic roles. In their seminal proposal, Gildea and Jurafsky approached the task using various features such as headword, phrase type, and parse tree path [6]. Such features have remained the basic and essential features in subsequent research, irrespective of the variation in the actual learning components. In addition, parsed sentences are often required, for extracting the path features during training and providing the argument boundaries during testing. The parse information is deemed important for the per- formance of role labelling [7, 8]. More precisely, in semantic role labelling, parse information is rather more critical for the identification of boundaries for candidate constituents than for the extraction Semantic Role Tagging for Chinese at the Lexical Level 805 of training data. Its limited function in training, for instance, is reflected in the low coverage reported (e.g. [21]). However, given the imperfection of existing automatic parsers, which are far from producing gold standard parses, many thus resort to shal- low syntactic information from simple chunking, though …

Content truncated to 500 words