Automatic verb classification using VerbNet-style

Abstract

The manual building of computational lexical resources is impracticable, mainly because of the hard work and its time consuming. The Natural Language Process area intends to facilitate this task by using automatic and semiautomatic computational methods to build these resources. One approach uses machine learning from corpus. Another uses a cross-linguistic approach by using existing computational lexical resources to build a new resource. In this project we will exploit clustering techniques to find syntactic-semantic verbal classes for the Brazilian Portuguese language - according to the first approach. Specifically, we intend to follow de VerbNet-style to perform verb clustering. VerbNet is a verb lexicon with syntactic and semantic information about English verbs, domain-independent, based on Levin's verb classes and with mappings to the Princeton WordNet. The results of this project will be compared with the results of VerbNet.Br (Master project of this student). The VerbNet.Br is being built by using the mappings among existing computational lexical resources: VerbNet, WordNet and WordNet.Br, using a semiautomatic method - according to the cross-linguistic approach. The aim of the comparison to be made is to verify if the semiautomatic method (more expensive) presents more accurate results than the clustering method (cheaper). (AU)