Parsing Syntactic and Semantic Dependencies with Two Single-Stage Maximum Entropy Models

CoNLL 2008: Proceedings of the 12th Conference on Computational Natural Language Learning, pages 203–207 Manchester, August 2008 Parsing Syntactic and Semantic Dependencies with Two Single-Stage Maximum Entropy Models ∗ Hai Zhao and Chunyu Kit Department of Chinese, Translation and Linguistics City University of Hong Kong 83 Tat Chee Avenue, Kowloon, Hong Kong, China haizhao@cityu.edu.hk, ctckit@cityu.edu.hk Abstract This paper describes our system to carry out the joint parsing of syntactic and se- mantic dependencies for our participation in the shared task of CoNLL-2008. We il- lustrate that both syntactic parsing and se- mantic parsing can be transformed into a word-pair classification problem and im- plemented as a single-stage system with the aid of maximum entropy modeling. Our system ranks the fourth in the closed track for the task with the following per- formance on the WSJ+Brown test set: 81.44% labeled macro F1 for the overall task, 86.66% labeled attachment for syn- tactic dependencies, and 76.16% labeled F1 for semantic dependencies. 1 Introduction The joint parsing of syntactic and semantic depen- dencies introduced by the shared task of CoNLL- 08 is more complicated than syntactic dependency parsing or semantic role labeling alone (Surdeanu et al., 2008). For semantic parsing, in particu- lar, a dependency-based representation is given but the predicates involved are unknown, and we also have nominal predicates besides the verbal ones. All these bring about more difficulties for learning. This paper presents our research for participation in the CoNLL-2008 shared task, with a highlight on our strategy to select learning framework and features for maximum entropy learning. ∗This study is supported by CERG grant 9040861 (CityU 1318/03H) and CityU Strategic Research Grant 7002037. ∗c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. The rest of the paper is organized as follows. The next section presents the technical details of our system and Section 3 its evaluation results. Section 4 looks into a few issues concerning our forthcoming work for this shared task, and Section 5 concludes the paper. 2 System Description For the sake of efficiency, we opt for the maximum entropy model with Gaussian prior as our learning model for both the syntactic and semantic depen- dency parsing. Our implementation of the model adopts L-BFGS algorithm for parameter optimiza- tion as usual (Liu and Nocedal, 1989). No addi- tional feature selection techniques are applied. Our system consists of three components to deal with syntactic and semantic dependency parsing and word sense determination, respectively. Both parsing is formulated as a single-stage word-pair classification problem, and the latter is carried out by a search through the NomBank (Meyers et al., 2004) or the PropBank (Palmer et al., 2005)1. 2.1 Syntactic Dependency Parsing We use a shift-reduce scheme to implement syn- tactic dependency parsing as in (Nivre, 2003). It takes a step-wised, history- or transition-based ap- proach. It is basically a word-by-word method with a projective constraint. In each step, the clas- sifier checks a word pair, e.g., TOP, the top of a stack …

Content truncated to 500 words