Ensemble Methods for Unsupervised WSD

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 97–104, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Ensemble Methods for Unsupervised WSD Samuel Brody School of Informatics University of Edinburgh s.brody@sms.ed.ac.uk Roberto Navigli Dipartimento di Informatica Universit`a di Roma “La Sapienza” navigli@di.uniroma1.it Mirella Lapata School of Informatics University of Edinburgh mlap@inf.ed.ac.uk Abstract Combination methods are an effective way of improving system performance. This paper examines the benefits of system combination for unsupervised WSD. We investigate several voting- and arbiter- based combination strategies over a di- verse pool of unsupervised WSD systems. Our combination methods rely on predom- inant senses which are derived automati- cally from raw text. Experiments using the SemCor and Senseval-3 data sets demon- strate that our ensembles yield signifi- cantly better results when compared with state-of-the-art. 1 Introduction Word sense disambiguation (WSD), the task of identifying the intended meanings (senses) of words in context, holds promise for many NLP applications requiring broad-coverage language understanding. Examples include summarization, question answering, and text simplification. Re- cent studies have also shown that WSD can ben- efit machine translation (Vickrey et al., 2005) and information retrieval (Stokoe, 2005). Given the potential of WSD for many NLP tasks, much work has focused on the computa- tional treatment of sense ambiguity, primarily us- ing data-driven methods. Most accurate WSD sys- tems to date are supervised and rely on the avail- ability of training data, i.e., corpus occurrences of ambiguous words marked up with labels indicat- ing the appropriate sense given the context (see Mihalcea and Edmonds 2004 and the references therein). A classifier automatically learns disam- biguation cues from these hand-labeled examples. Although supervised methods typically achieve better performance than unsupervised alternatives, their applicability is limited to those words for which sense labeled data exists, and their accu- racy is strongly correlated with the amount of la- beled data available (Yarowsky and Florian, 2002). Furthermore, obtaining manually labeled corpora with word senses is costly and the task must be repeated for new domains, languages, or sense in- ventories. Ng (1997) estimates that a high accu- racy domain independent system for WSD would probably need a corpus of about 3.2 million sense tagged words. At a throughput of one word per minute (Edmonds, 2000), this would require about 27 person-years of human annotation effort. This paper focuses on unsupervised methods which we argue are useful for broad coverage sense disambiguation. Unsupervised WSD algo- rithms fall into two general classes: those that per- form token-based WSD by exploiting the simi- larity or relatedness between an ambiguous word and its context (e.g., Lesk 1986); and those that perform type-based WSD, simply by assigning all instances of an ambiguous word its most fre- quent (i.e., predominant) sense (e.g., McCarthy et al. 2004; Galley and McKeown 2003). The pre- dominant senses are automatically acquired from raw text without recourse to manually annotated data. The motivation for assigning all instances of a word to its most prevalent sense stems from …

Content truncated to 500 words