MELB-YB: Preposition Sense Disambiguation Using Rich Semantic Features

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 241–244, Prague, June 2007. c⃝2007 Association for Computational Linguistics MELB-YB: Preposition Sense Disambiguation Using Rich Semantic Features Patrick Ye and Timothy Baldwin Computer Science and Software Engineering University of Melbourne, Australia {jingy,tim}@csse.unimelb.edu.au Abstract This paper describes a maxent-based prepo- sition sense disambiguation system entry to the preposition sense disambiguation task of the SemEval 2007. This system uses a wide variety of semantic and syntactic fea- tures to perform the disambiguation task and achieves a precision of 69.3% over the test data. 1 Introduction Prepositional phrases (PPs) are both common and semantically varied in open English text. While the conventional view on prepositions from the com- putational linguistics community has been that they are semantically transient at best, and semantically- vacuous at worst, a robust account of the semantics of prepositions and disambiguation method can be helpful in a range of NLP tasks including machine translation, parsing (prepositional phrase attach- ment) and semantic role labelling (Durand, 1993; O’Hara and Wiebe, 2003; Ye and Baldwin, 2006a). The SemEval 2007 preposition sense disambigua- tion task provides a common test bed for the evalua- tion of preposition sense disambiguation systems. Our proposed method is maximum entropy based, and combines features developed in the context of preposition sense disambiguation for semantic role labelling (Ye and Baldwin, 2006a), and verb sense disambiguation (Ye and Baldwin, 2006b). The remainder of this paper is structured as fol- lows. We first discuss the pre-processing steps used in our system (Section 2), and outline the fea- tures our preposition disambiguation method uses (Section 3) and our parameter tuning method (Sec- tion 4). We then discuss and analyse the results of our method (Section 5) and conclude the paper (Sec- tion 6). 2 Pre-processing The following list shows the pre-processing steps that our system goes through and the tools used: Part of speech tagging SVMTool version 1.2 (Gim´enez and M`arquez, 2004). Chunking An in-house chunker implemented with fnTBL, a transformation based learner (Ngai and Florian, 2001), and trained on the British Na- tional Corpus (BNC).1 Parsing Charniak’s re-ranking parser, version Au- gust, 2006 (Charniak and Johnson, 2005). Named entity extraction A statistical NER sys- tem described in Cohn et al. (2005). Supersense tagging A WordNet-based super- sense tagger (Ciaramita and Altun, 2006). Semantic role labeling ASSERT version 1.4 (Pradhan et al., 2004). 3 Features The disambiguation features used by our system can be divided into three categories: collocation fea- tures, syntactic features and semantic-role based fea- tures. We discuss each in turn below. 3.1 Collocation Features The collocation features were inspired by the one-sense-per-collocation heuristic proposed by Yarowsky (1995). These features were designed to capture open class words that exhibit strong colloca- tion properties with respect to the different senses of the target preposition. Details of the features in this category are listed below. 1This chunker is not exactly the same as Ngai and Florian’s system, however it does use the default transformation tem- plates supplied by fnTBL. 241 Bag …

Content truncated to 500 words