Preprint version of the paper published in Expert Systems with Applications, doi:10.1016/j.eswa.2011.01.150 Understanding the Role of Conceptual Relations in Word Sense Disambiguation David Fernandez-Amoros Dpto. de Lenguajes y Sistemas Informaticos david@lsi.uned.es Ruben Heradio Dpto. de Ingenieria de Software y Sistemas Informaticos rheradio@issi.uned.es Universidad Nacional de Educacion a Distancia Madrid, Spain Abstract In this article, we concentrate in conceptual relations as a source of information for Word Sense Dis- ambiguation (WSD) systems. We start with a review the most relevant research in the field, then we implement our own algorithm. As a starting point we have chosen the conceptual density algorithm of Agirre and Rigau. We generalize the original algorithm, parameterizing many aspects. This new algo- rithm obtains a relative improvement of 24% in terms of precision and recall. We also offer comparative evaluation of our system with respect to the participants in the SENSEVAL-2 disambiguation competition. We conclude that conceptual relations provide a source of information that is insufficient by itself to achieve good disambiguation results, but can, however, be a very accurate heuristic in a combined system. 1 Introduction Word Sense Disambiguation (WSD) is the problem of determining, for a word in context, the specific mean- ing of the word in a dictionary, or, more generally, a sense inventory. The classical example would the word bank in sentences like I asked for a loan in the bank or I sat in the bank of the river. There are several ways to evaluate the performance of a WSD disambiguation system. Precision is informally the accuracy of the system over the words it has been able to disambiguate. Recall is the measure of the performance of the system overall. With the appropriate definition, recall = precision · coverage. Coverage is the ratio of disambiguated words (correctly or not) over the total number of words. These measures are often computed comparing the results of a system with a gold standard that has been disambiguated by hand. WordNet [Miller, 1995] is a lexical knowledge base that is also a popular sense inventory. The SemCor collection [Francis and Kucera, 1967] is a subset of the Brown Corpus tagged by hand. Much work in WSD has used WordNet and SemCor as sense inventory and gold standard respectively, for instance [Agirre and Martinez, 2001, Agirre and Rigau, 1995, Agirre and Rigau, 1996, Voorhees, 1993, Sussna, 1993,Chodorow et al., 2000,Cucchiarelli et al., 2000,Dini et al., 1998,Dorr and Jones, 1996,Fellbaum et al., 1997, Haynes, 2001, Krovetz, 1998, Kwong, 2001, Lin, 1997, Mihalcea and Moldovan, 2000a, Mihalcea 1