Verb-Particle Constructions in the World Wide Web

In this paper we investigate the phenomenon of verb-particle constructions, discussing their characteristics and their availability for use with NLP systems. Combinations automatically extracted from corpora greatly improve the coverage of available resources. However, the data sparseness problem is particularly acute for these constructions and even using a corpus as large as the British National Corpus, a great proportion of combinations have a very low frequency, while others never occur in it. In this paper we propose using the World Wide Web as a way to validate candidate combinations minimising the problem of data sparseness. This method can be use to extend the coverage of existing lexical resources by validating combinations automatically generated from classes of verbs, and to improve the reliability of those combinations automatically extracted from corpora.