Kyoto: An Integrated System for Specific Domain WSD

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 417–420, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics Kyoto: An Integrated System for Specific Domain WSD Aitor Soroa, Eneko Agirre, Oier Lopez de Lacalle University of the Basque Country a.soroa@ehu.es Monica Monachini Istituto di Linguistica Computazionale monica.monachini@ilc.cnr.it Jessie Lo, Shu-Kai Hsieh National Taiwan Normal University shukai@ntnu.edu.tw Wauter Bosma, Piek Vossen Vrije Universiteit {p.vossen,w.bosma}@let.vu.nl Abstract This document describes the prelimi- nary release of the integrated Kyoto sys- tem for specific domain WSD. The sys- tem uses concept miners (Tybots) to ex- tract domain-related terms and produces a domain-related thesaurus, followed by knowledge-based WSD based on word- net graphs (UKB). The resulting system can be applied to any language with a lexical knowledge base, and is based on publicly available software and resources. Our participation in Semeval task #17 fo- cused on producing running systems for all languages in the task, and we attained good results in all except Chinese. Due to the pressure of the time-constraints in the competition, the system is still under development, and we expect results to im- prove in the near future. 1 Introduction In this paper we describe the participation of the integrated Kyoto system on the “SemEval-2010 task #17: All-words Word Sense Disambigua- tion on a Specific Domain” task (Agirre et al., 2010). The goal of our participation was to eval- uate the preliminary release of the integrated sys- tem for specific domain WSD developed for the Kyoto project1. Besides, we wanted to test the performance of our domain specific WSD system (Agirre et al., 2009) on this test set, and to inte- grate the thesaurus construction software (Tybots) developed for the project. The system can be run for any language and domain if provided with a lexical knowledge base and some background doc- uments on the domain. We will first present the components of our sys- tem, followed by the experimental design and the 1http://www.kyoto-project.eu results. Finally, the conclusions are presented. 2 The Kyoto System for Domain Specific WSD We will present in turn UKB, the Tybots, and the lexical knowledge-bases used. 2.1 UKB UKB is a knowledge-based unsupervised WSD system which exploits the structure of an under- lying Language Knowledge Base (LKB) and finds the most relevant concepts given an input con- text (Agirre and Soroa, 2009). UKB starts by tak- ing the LKB as a graph of concepts G = (V, E) with a set of vertices V derived from LKB con- cepts and a set of edges E representing relations among them. Giving an input context, UKB ap- plies the so called Personalized PageRank (Haveli- wala, 2002) over it to obtain the most representa- tive senses for the context. PageRank (Brin and Page, 1998) is a method for scoring the vertices V of a graph according to each node’s structural importance. The algo- rithm can be viewed as random walk process that postulate the existence of a particle that randomly traverses the graph, but at any …

Content truncated to 500 words