Automatic Detection of Non-deverbal Event Nouns for Quick Lexicon Production

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 46–52, Beijing, August 2010 Automatic Detection of �on-deverbal Event �ouns for Quick Lexicon Production �úria Bel IULA Universitat Pompeu Fabra nuria.bel@upf.edu Maria Coll IULA Universitat Pompeu Fabra maria.coll02@ campus.upf.edu Gabriela Resnik Universidad Nacional de General Sarmiento and Uni- versitat Pompeu Fabra gresnik@ungs.edu.ar Abstract In this work we present the results of ex- perimental work on the development of lexical class-based lexica by automatic means. Our purpose is to assess the use of linguistic lexical-class based informa- tion as a feature selection methodology for the use of classifiers in quick lexical development. The results show that the approach can help reduce the human ef- fort required in the development of lan- guage resources significantly. 1 Introduction Although language independent, many linguistic technologies are inherently tied to the availabili- ty of particular language data (i.e. Language Re- sources, LR). The nature of these data is very much dependent on particular technologies and the applications where are used. Currently, most systems are using LR collected by hand that still do not cover all languages, or all possible appli- cation domains, or all possible information re- quired by the many applications that are being proposed. Methods for the automatic and quick development of new LR have to be developed in order to guarantee a supply of the required data. Esuli and Sebastiani (2006) did a classification experiment for creating lexica for opinion min- ing, for instance, and the importance of lexical information for event extraction in Biomedical texts has been addressed in Fillmore et al. (2006). One way of producing such resources is to classify words into lexical classes via methods based on their morphosyntactic contexts of oc- currence. In the next three sections we report on an ex- periment on cue-based lexical classification for non-deverbal event nouns, that is, nouns such as ‘party’ or ‘conflict’, which refer to an event but cannot be identified by their morphology, as is the case with deverbal nouns such as ‘construc- tion’. The purpose of this experiment was, as already stated, to investigate methods for the rapid generation of an event nouns lexicon for two different languages, using a reduced quanti- ty of available texts. Assuming that linguistic information can be provided by occurrence dis- tribution, as is usually done in linguistic theory to motivate lexical classes (e.g. Grimshaw, 1990), cue information has been gathered from texts and used to train and test a Decision Tree- based classifier. We experimented with two dif- ferent languages to test the potential coverage of the proposed technique in terms of its adaptation to different languages, and also used different types of corpora to test its adaptability to differ- ent domains and sizes. 2 Some properties of �on-Deverbal Event �ouns in Spanish and English. We based our experiment on the work by Resnik (2004) who proposes a specific lexical class for Spanish event nouns like accidente (‘accident’) or guerra (‘war’) which cannot be identified by suffixes such as ‘-ción’ …

Content truncated to 500 words