UCD-FC: Deducing semantic relations using WordNet senses that occur frequently in a database of noun-noun compounds

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 370–373, Prague, June 2007. c⃝2007 Association for Computational Linguistics UCD-FC: Deducing semantic relations using WordNet senses that occur frequently in a database of noun-noun compounds ∗ Fintan J. Costello, School of Computer Science and Informatics, University College Dublin, Belfield, Dublin 6, Ireland. fintan.costello@ucd.ie Abstract This paper describes a system for classify- ing semantic relations among nominals, as in SemEval task 4. This system uses a corpus of 2,500 compounds annotated with WordNet senses and covering 139 different semantic relations. Given a set of nomi- nal pairs for training, as provided in the Se- mEval task 4 training data, this system con- structs for each training pair a set of features made up of relations and WordNet sense pairs which occurred with those nominals in the corpus. A Naive Bayes learning al- gorithm learns associations between these features and relation membership categories. The identification of relations among nomi- nals in test items takes place on the basis of these associations. 1 Introduction This paper describes a system for deducing the correct semantic relation between a pair of nom- inals in a sentence, as in SemEval task 4 (Girju, Hearst, Nakov, Nastase, Szpakowicz, Turney, & Yuret, 2007). This system is an adaptation of an existing system for deducing the correct semantic relation between the pair of words in a noun-noun compound. This compound disambiguation system (named PRO, for Proportional Relation Occurrence; see Costello, Veale, & Dunne, 2006) makes use of ∗ This research was supported by the FP6 NEST Pro- gramme of the European Commission (ANALOGY: Humans the Analogy-Making Species: STREP Contr. No 029088) a corpus of 2,500 compounds annotated with Word- Net senses and covering 139 different semantic re- lations, with each noun and each relation annotated with its correct WordNet sense.1 Section 2 of the pa- per will describe the format and structure of this cor- pus, Section 3 will describe the original PRO com- pound disambiguation system, and Section 4 will explain how the PRO system was adapted to deduce the correct semantic relation between a pair of nom- inals, as in SemEval task 4. Four different versions of the adapted system were produced (versions A,B, C and D), either using or not using the WordNet la- bels and the Query labels provided with training and test items in SemEval task 4. Section 5 discusses the performance of these different versions of the sys- tem. Finally, Section 6 finishes the paper with some discussion and ideas for future work. 2 A Corpus of Annotated Compounds Using WordNet (Miller, 1995), version 2.0, a cor- pus of noun-noun compounds was constructed such that each compound was annotated with the correct WordNet noun senses for constituent words, the cor- rect semantic relation between those words, and the correct WordNet verb sense for that relation, as de- scribed below. 2.1 Corpus Procedure The compounds used in this corpus were selected from the set of noun-noun compounds defined in WordNet. Compounds from WordNet …

Content truncated to 500 words