A Dependency Treebank of the Chinese Buddhist Canon

A Dependency Treebank of the Chinese Buddhist Canon Tak-sum Wong, John Lee Halliday Centre for Intelligent Applications of Language Studies Department of Linguistics and Translation, City University of Hong Kong E-mail: tswong-c@my.cityu.edu.hk, jsylee@cityu.edu.hk Abstract We present a dependency treebank of the Chinese Buddhist Canon, which contains 1,514 texts with about 50 million Chinese characters. The treebank was created by an automatic parser trained on a smaller treebank, containing four manually annotated sutras (Lee and Kong, 2014). We report results on word segmentation, part-of-speech tagging and dependency parsing, and discuss challenges posed by the processing of medieval Chinese. In a case study, we exploit the treebank to examine verbs frequently associated with Buddha, and to analyze usage patterns of quotative verbs in direct speech. Our results suggest that certain quotative verbs imply status differences between the speaker and the listener. Keywords: dependency treebank; Chinese Buddhist canon; quotative verbs 1. Introduction Over the past decade, there has been growing interest in building treebanks for historical texts, not only for facilitating their reading but also for studying the historical languages in which they were written. The sacred texts of many major religious, for example, have been syntactically analysed: treebanks are now available for the Hebrew Bible (Wu & Lowery, 2006), the New Testament in Greek (Haug & Jøhndal, 2008), and the Qur’an in Classical Arabic (Dukes & Buckwalter, 2010). With about 50 million characters, the sheer volume of the Chinese Buddhist Canon makes it difficult for any individual to perform manual analysis over the entire corpus. Although digitized versions of the Canon have enabled n-gram and other lexical analyses (Lancaster, 2010), it remains difficult to examine patterns in part-of-speech (POS) and sentence structures without syntactic annotations. To date, the only treebank with Buddhist Chinese material consists only of four sutras (Lee and Kong, 2014). We trained a dependency parser on this small treebank, and then automatically parsed the entire Chinese Buddhist Canon. In this paper, we start with an overview of existing treebanks for ancient Chinese (Section 2). We then report the procedure for constructing this treebank, and evaluate its accuracy (Section 3). Finally, as a case study, we examine the verbs used by Buddha and other characters in the treebank, focusing on quotative verbs (Section 4). 2. Previous Work Among large, diachronic corpora in ancient Chinese are the Academia Sinica Ancient Chinese Corpus (Wei et al., 1997) and the Sheffield Corpus of Chinese (Hu et al., 2005), both covering a wide range of time and genres. Although word-segmented and POS-tagged, they have not been syntactically analyzed. To the best of our knowledge, only three treebanks are available to-date for ancient Chinese. First, a constituent-based treebank has been constructed on 1,000 sentences from pre-Qin texts (Huang et al., 2002). Second, a dependency treebank has been annotated for 32,000 characters of Tang poems, selected from the works of three poets from the 8th century CE (Lee and Kong, 2012). Third, and most related to this work, is a dependency treebank of four sutras, numbering …

Content truncated to 500 words