Building A Case-based Semantic English-Chinese Parallel Treebank

Building a Case-based Semantic English-Chinese Parallel Treebank Huaxing Shi1, Tiejun Zhao1, Keh-Yih Su2 1Harbin Institute of Technology (HIT), 2Academia Sinica 1 School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China 2128 Academia Road, Section 2, Nankang, Taipei 11529, Taiwan 1{hxshi, tjzhao}@mtlab.hit.edu.cn, 2kysu@iis.sinica.edu.tw Abstract We construct a case-based English-to-Chinese semantic constituent parallel Treebank for a Statistical Machine Translation (SMT) task by labelling each node of the Deep Syntactic Tree (DST) with our refined semantic cases. Since subtree span-crossing is harmful in tree-based SMT, DST is adopted to alleviate this problem. At the same time, we tailor an existing case set to represent bilingual shallow semantic relations more precisely. This Treebank is a part of a semantic corpus building project, which aims to build a semantic bilingual corpus annotated with syntactic, semantic cases and word senses. Data in our Treebank is from the news domain of Datum corpus. 4,000 sentence pairs are selected to cover various lexicons and part-of-speech (POS) n-gram patterns as much as possible. This paper presents the construction of this case Treebank. Also, we have tested the effect of adopting DST structure in alleviating subtree span-crossing. Our preliminary analysis shows that the compatibility between Chinese and English trees can be significantly increased by transforming the parse-tree into the DST. Furthermore, the human agreement rate in annotation is found to be acceptable (90% for English nodes, 75% for Chinese nodes). Keywords: English-Chinese semantic constituent parallel Treebank, case tree annotation, semantic machine translation corpus 1. Introduction The reason of building this corpus is that having a bilingual semantic corpus with refined semantic role information is expected to bring significant benefit to the task of Statistical Machine Translation (SMT). Since the semantic constituent is less variant during translation in comparison with the syntactic constituent (Fung et al., 2007), it should lessen the data sparseness problem of translation patterns, which often occur in syntactic SMT. Researchers have paid attention to constructing se- mantic resources in the last 20 years. And many useful and high-quality semantic resources have been built, such as FrameNet (Baker et al., 1998), PropBank (Palmer et al., 2005), Academia Sinica Treebank (Huang et al., 2000), NomBank (Meyers et al., 2004) , VerbNet (Schuler, 2005), HowNet (Dong and Dong, 2003), etc. However, each of them only serves its own purpose. Therefore, they are different from each other in many details, such as annotation types (e.g., frame sets in FrameNet, shared semantic arguments in PropBank) and annotation methods (e.g., adding a layer of predicate-argument information to syntactic structures in PropBank, and labeling semantic information on each node as Sinica Treebank does). Currently, FrameNet and PropBank are the two most commonly used semantic resources in Semantic Role Labeling (SRL) and SMT tasks. FrameNet is based on a theory of meaning called Frame Semantics (Fillmore, 1982). The basic idea is that the meanings of most words can best be understood on the basis of their semantic frames, which describe the types of events, relations, and the participants involved. In FrameNet, most …

Content truncated to 500 words