UBC-UMB: Combining unsupervised and supervised systems for all-words WSD

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 350–353, Prague, June 2007. c⃝2007 Association for Computational Linguistics UBC-UMB: Combining unsupervised and supervised systems for all-words WSD David Martinez,Timothy Baldwin LT Group, CSSE University of Melbourne Victoria 3010 Australia {davidm,tim}@csse.unimelb.edu.au Eneko Agirre, Oier Lopez de Lacalle IXA NLP Group Univ. of the Basque Country Donostia, Basque Country {e.agirre,jibloleo}@ehu.es Abstract This paper describes the joint submission of two systems to the all-words WSD sub- task of SemEval-2007 task 17. The main goal of this work was to build a competitive unsupervised system by combining hetero- geneous algorithms. As a secondary goal, we explored the integration of unsupervised predictions into a supervised system by dif- ferent means. 1 Introduction This paper describes the joint submission of two sys- tems to the all-words WSD subtask of SemEval- 2007 task 17. The systems were developed by the University of the Basque Country (UBC), and the University of Melbourne (UMB). The main goal of this work was to build a competitive unsupervised system by combining heterogeneous algorithms. As a secondary goal, we explored the integration of this method into a supervised system by different means. Thus, this paper describes both the unsu- pervised system (UBC-UMB-1), and the combined supervised system (UBC-UMB-2) submitted to the all-words task. Our motivation in building unsupervised systems comes from the difficulty of creating hand-tagged data for all words and all languages, which is col- loquially known as the knowledge acquisition bot- tleneck. There have also been promising results in recent work on the combination of unsupervised ap- proaches that suggest the gap with respect to super- vised systems is narrowing (Brody et al., 2006). The remainder of the paper is organized as fol- lows. First we describe the disambiguation algo- rithms in Section 2. Next, the development exper- iments are presented in Section 3, and our final sub- missions and results in Section 4. Finally, we sum- marize our conclusions in Section 5. 2 Algorithms In this section, we will describe the standalone algo- rithms (three unsupervised and one supervised) and the combination schemes we explored. The unsu- pervised methods are based on different intuitions for disambiguation (topical features, local context, and WordNet relations), which is a desirable charac- teristic for combining algorithms. 2.1 Topic Signatures (TS) Topic signatures (Agirre and de Lacalle, 2004) are lists of words related to a particular sense. They can be built from a variety of sources, and be used di- rectly to perform WSD. Cuadros and Rigau (2006) present a detailed evaluation of topic signatures built from a variety of knowledge sources. In this work we built those coming from the following: • the relations in the Multilingual Central Repos- itory (TS-MCR) • the relations in the Extended WordNet (TS- XWN) In order to apply this resource for WSD, we sim- ply measured the word-overlap between the target context and each of the senses of the target word. The sense with highest overlap is chosen as the cor- rect sense. 350 2.2 …

Content truncated to 500 words