22. Extracting specialized collocations using lexical functions

This chapter presents a method for extracting specialized verb + noun collocations based on lexical functions, the formal device developed within the framework of theMeaning-Text theory to represent collocations. Although based on a linguistic analysis of collocations, our program (Colex) uses statistical measures to distinguish true collocations from free combinations.We argue that log-likelihood is a better measure of precision in this domain than mutual information.We are now developing the semantic encoding of the combinations extracted by Colex. We believe that a filter based on measuring a combination’s ability to be encoded as a lexical function would achieve higher precision than a purely statistical filter.