CUNIT: A Semantic Role Labeling System for Modern Standard Arabic

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 133–136, Prague, June 2007. c⃝2007 Association for Computational Linguistics CUNIT: A Semantic Role Labeling System for Modern Standard Arabic Mona Diab Columbia University mdiab@cs.columbia.edu Alessandro Moschitti University of Trento, DIT moschitti@dit.unitn.it Daniele Pighin FBK-irst; University of Trento, DIT pighin@itc.it Abstract In this paper, we present a system for Ara- bic semantic role labeling (SRL) based on SVMs and standard features. The system is evaluated on the released SEMEVAL 2007 development and test data. The results show an Fβ=1 score of 94.06 on argument bound- ary detection and an overall Fβ=1 score of 81.43 on the complete semantic role label- ing task using gold parse trees. 1 Introduction There is a widely held belief in the computational linguistics field that identifying and defining the roles of predicate arguments, semantic role label- ing (SRL), in a sentence has a lot of potential for and is a significant step towards the improvement of important applications such as document retrieval, machine translation, question answering and infor- mation extraction. However, effective ways for see- ing this belief come to fruition require a lot more research investment. Since most of the available data resources are for the English language, most of the reported SRL sys- tems to date only deal with English. Nevertheless, we do see some headway for other languages, such as German and Chinese (Erk and Pado, 2006; Sun and Jurafsky, 2004; Xue and Palmer, 2005). The systems for non-English languages follow the suc- cessful models devised for English, e.g. (Gildea and Jurafsky, 2002; Xue and Palmer, 2004; Pradhan et al., 2003). However, no SRL system exists for Ara- bic. In this paper, we present a system for semantic role labeling for modern standard Arabic. To our knowledge, it is the first SRL system for a semitic language in the literature. It is based on a supervised model that uses support vector machines (SVM) technology for argument boundary detection and ar- gument classification. It is trained and tested using the pilot Arabic PropBank data released as part of the SEMEVAL 2007 data. Given the lack of a re- liable deep syntactic parser, in this research we use gold trees. The system yields an F-score of 94.06 on the sub task of argument boundary detection and an F-score of 81.43 on the complete task, i.e. boundary plus classification. 2 SRL system for Arabic The design of an optimal model for an Arabic SRL systems should take into account specific linguis- tic aspects of the language. However, a remarkable amount of research has already been done in SRL and we can capitalize from it to design a basic and effective SRL system. The idea is to use the technol- ogy developed for English and verify if it is suitable for Arabic. Our adopted SRL models use Support Vector Ma- chines (SVM) to implement a two steps classifica- tion approach, i.e. boundary detection and argument classification. Such models have already been in- vestigated in (Pradhan et …

Content truncated to 500 words