Semantic Role Labeling Using libSVM

Proceedings of the 9th Conference on Computational Natural Language Learning (CoNLL), pages 205–208, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Semantic Role Labeling using libSVM Necati Ercan Ozgencil Nancy McCracken Center for Natural Language Processing Center for Natural Language Processing School of Engineering and Computer Science School of Information Studies Syracuse University Syracuse University neozgenc@ecs.syr.edu njm@ecs.syr.edu Abstract We describe a system for the CoNLL- 2005 shared task of Semantic Role Label- ing. The system implements a two-layer architecture to first identify the arguments and then to label them for each predicate. The components are implemented as SVM classifiers using libSVM. Features were adapted and tuned for the system, including a reduced set for the identifier classifier. Experiments were conducted to find kernel parameters for the Radial Ba- sis Function (RBF) kernel. An algorithm was defined to combine the results of the argument labeling classifier according to the constraints of the argument labeling problem. 1 Introduction and Strategy The Semantic Role Labeling (SRL) problem has been the topic of the both the CoNLL-2004 and the CoNLL-2005 Shared Tasks (Carreras and Màrquez, 2005). The SRL system described here depends on a full syntactic parse from the Charniak parser, and investigates aspects of using Support Vector Machines (SVMs) as the machine learning technique for the SRL problem, using the libSVM package. In common with many other systems, this sys- tem uses the two-level strategy of first identifying which phrases can be arguments to predicates in general, and then labeling the arguments according to that predicate. The argument identification phase is a binary classifier that decides whether each constituent in the full syntax tree of the sen- tence is a potential argument. These potential ar- guments are passed into the argument labeling classifier, which uses binary classifiers for each label to decide if that label should be given to that argument. A post-processing phase picks the best labeling that satisfies the constraints of labeling the predicate arguments. For overall classification strategy and for suggestions of features, we are indebted to the work of Pradhan et al (2005) and to the work of many authors in both the CoNLL-2004 shared task and the similar semantic roles task of Senseval-3. We used the results of their experiments with features, and worked primarily on features for the identifying classifier and with the constraint satisfaction problem on the final argument output. 2 System Description 2.1 Input Data In this system, we chose to use full syntax trees from the Charniak parser, as the constituents of those trees more accurately represented argument phrases in the training data at the time of the data release. Within each sentence, we first map the predicate to a constituent in the syntax tree. In the cases that the predicate is not represented by a con- stituent, we found that these were verb phrases of length two or more, where the first word was the main verb (carry out, gotten away, served up, etc.). In these cases, we used the first word …

Content truncated to 500 words