SemEval-2012 Task 5: Chinese Semantic Dependency Parsing

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 378–384, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics SemEval-2012 Task 5: Chinese Semantic Dependency Parsing Wanxiang Che†, Meishan Zhang†, Yanqiu Shao‡, Ting Liu† †Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China {car, mszhang, tliu}@ir.hit.edu.cn ‡Beijing City University, China yqshao@bcu.edu.cn Abstract The paper presents the SemEval-2012 Shared Task 5: Chinese Semantic Dependency Pars- ing. The goal of this task is to identify the de- pendency structure of Chinese sentences from the semantic view. We firstly introduce the motivation of providing Chinese semantic de- pendency parsing task, and then describe the task in detail including data preparation, data format, task evaluation, and so on. Over ten thousand sentences were labeled for partici- pants to train and evaluate their systems. At last, we briefly describe the submitted systems and analyze these results. 1 Introduction Semantic analysis is a long-term goal of Natural Language Processing, and as such, has been re- searched for several decades. A number of tasks for encoding semantic information have been devel- oped over the years, such as entity type recognition and word sense disambiguation. Recently, sentence- level semantics – in particular, semantic role label- ing – has received increasing attention. However, some problems concerning the semantic representa- tion method used in semantic role labeling continue to exist (Xue and Palmer, 2005). 1. Semantic role labeling only considers predicate-argument relations and ignores the semantic relations between a noun and its modifier. 2. The meaning of semantic roles is related to spe- cial predicates. Therefore, there are infinite se- mantic roles to be learned, as the number of predicates is not fixed. Although the Prop- Bank (Xue and Palmer, 2003) normalizes these semantic roles into certain symbols, such as Arg0-Arg5, the same symbol can have different semantic meanings when paired with different predicates, and thus cannot be learned well. Semantic dependency parsing is therefore pro- posed to solve the two problems above for Chinese. Firstly, the proposed method analyzes all the words’ semantic roles in a sentence and specifies the con- crete semantic relation of each word pair. After- ward, this work analyzes and summarizes all the possible semantic roles, obtaining over 100 of them, and then uses these semantic roles to specify the se- mantic relation for each word pair. Dependency parsing (K¨ubler et al., 2009) is based on dependency grammar. It has several advantages, such as concise formalization, easy comprehension, high efficiency, and so on. Dependency parsing has been studied intensively in recent decades, with most related work focusing on syntactic structure. Many research papers on Chinese linguistics demon- strate the remarkable difference between semantics and syntax (Jin, 2001; Zhou and Zhang, 2003). Chinese is a meaning-combined language with very flexible syntax, and semantics are more stable than syntax. The word is the basic unit of semantics, and the structure and meaning of a sentence consists mainly of a series of semantic dependencies between individual words (Li et al., 2003). Thus, …

Content truncated to 500 words