An Empirical Study on Class-Based Word Sense Disambiguation

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 389–397, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics An Empirical Study on Class-based Word Sense Disambiguation∗ Rub´en Izquierdo & Armando Su´arez Deparment of Software and Computing Systems University of Alicante. Spain {ruben,armando}@dlsi.ua.es German Rigau IXA NLP Group. EHU. Donostia, Spain german.rigau@ehu.es Abstract As empirically demonstrated by the last SensEval exercises, assigning the appro- priate meaning to words in context has re- sisted all attempts to be successfully ad- dressed. One possible reason could be the use of inappropriate set of meanings. In fact, WordNet has been used as a de-facto standard repository of meanings. How- ever, to our knowledge, the meanings rep- resented by WordNet have been only used for WSD at a very fine-grained sense level or at a very coarse-grained class level. We suspect that selecting the appropriate level of abstraction could be on between both levels. We use a very simple method for deriving a small set of appropriate mean- ings using basic structural properties of WordNet. We also empirically demon- strate that this automatically derived set of meanings groups senses into an adequate level of abstraction in order to perform class-based Word Sense Disambiguation, allowing accuracy figures over 80%. 1 Introduction Word Sense Disambiguation (WSD) is an inter- mediate Natural Language Processing (NLP) task which consists in assigning the correct semantic interpretation to ambiguous words in context. One of the most successful approaches in the last years is the supervised learning from examples, in which statistical or Machine Learning classification mod- els are induced from semantically annotated cor- pora (M`arquez et al., 2006). Generally, super- vised systems have obtained better results than the unsupervised ones, as shown by experimental work and international evaluation exercises such ∗This paper has been supported by the European Union under the projects QALL-ME (FP6 IST-033860) and KY- OTO (FP7 ICT-211423), and the Spanish Government under the project Text-Mess (TIN2006-15265-C06-01) and KNOW (TIN2006-15049-C03-01) as Senseval1. These annotated corpora are usu- ally manually tagged by lexicographers with word senses taken from a particular lexical semantic re- source –most commonly WordNet2 (WN) (Fell- baum, 1998). WN has been widely criticized for being a sense repository that often provides too fine–grained sense distinctions for higher level applications like Machine Translation or Question & Answer- ing. In fact, WSD at this level of granularity has resisted all attempts of inferring robust broad- coverage models. It seems that many word–sense distinctions are too subtle to be captured by auto- matic systems with the current small volumes of word–sense annotated examples. Possibly, build- ing class-based classifiers would allow to avoid the data sparseness problem of the word-based ap- proach. Recently, using WN as a sense reposi- tory, the organizers of the English all-words task at SensEval-3 reported an inter-annotation agree- ment of 72.5% (Snyder and Palmer, 2004). In- terestingly, this result is difficult to outperform by state-of-the-art sense-based WSD systems. Thus, some research has been focused on deriv- ing …

Content truncated to 500 words