PKU_HIT: An Event Detection System Based on Instances Expansion and Rich Syntactic Features

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 304–307, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics PKU_HIT: An Event Detection System Based on Instances Expansion and Rich Syntactic Features Shiqi Li1, Pengyuan Liu2, Tiejun Zhao1, Qin Lu3 and Hanjing Li1 1School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001, China {sqli,tjzhao,hjlee}@mtlab.hit.edu.cn 2Institute of Computational Linguistics, Peking University, Beijing 100871, China liupengyuan@pku.edu.cn 3Department of Computing, The Hong Kong Polytechnic University, Hong Kong, China csluqin@comp.polyu.edu.hk Abstract This paper describes the PKU_HIT system on event detection in the SemEval-2010 Task. We construct three modules for the three sub-tasks of this evaluation. For target verb WSD, we build a Naïve Bayesian classifier which uses additional training instances expanded from an untagged Chinese corpus automatically. For sentence SRL and event detection, we use a feature-based machine learning method which makes combined use of both constituent-based and dependency- based features. Experimental results show that the Macro Accuracy of the WSD module reaches 83.81% and F-Score of the SRL module is 55.71%. 1 Introduction In this paper, we describe the system submitted to the SemEval-2010 Task 11 on event detection in Chinese news sentences (Zhou, 2010). The objective of the task is to detect and analyze basic event contents in Chinese news sentences, similar to the frame semantic structure extraction task in SemEval-2007. However, this task is a more complex as it involves three interrelated subtasks: (1) target verb word sense disambiguation (WSD), (2) sentence semantic role labeling (SRL) and (3) event detection (ED). Therefore, the architecture of the system that we develop for the task consists of three modules: WSD, SRL and ED. First, the WSD module is to recognize key verbs or verb phrases which describe the basic event in a sentence, and then select an appropriate situation description formula for the recognized key verbs (or verb phrases); Then, the SRL module anchors the arguments to suitable constituents in the sentence, and then label each argument with three functional tags, namely constituent type tag, semantic role tags and event role tag. Finally, in the ED module, complete situation description of the sentence can be achieved by combining the results of the WSD module and the SRL module. For the WSD module, we consider the subtask as a general WSD problem. First of all, we automatically extract many instances from an untagged Chinese corpus using a heuristic rule inspired by Yarowsky (1993). Then we train a Naïve Bayesian (NB) classifier based on both the extracted instances and the official training data. We then use the NB classifier to predict situation the description formula and natural explanation of each target verb in testing data. For the SRL module, we use a rich syntactic feature-based learning method. As the state-of- the-art method in the field of SRL, feature-based method represents a predicate-argument structure (PAS) by a flat vector using a set of linguistic features. Then PAS can be directly classified by machine learning algorithms based on the corresponding …

Content truncated to 500 words