From machine readable dictionaries to lexicons for NLP: the cobuild dictionaries - a different approach

We describe the results of a syntactic-semantic parser for Cobuild dictionary definitions. Unlike previous work on the automatic analysis of machine readable dictionaries, the particular structure of the Cobuild definition allows us to derive information that classifies the lexical item mainly in terms of the selectional restrictions or preferences encoded on its arguments. The resulting formalized lexical entries contain data that has generally been lacking in other lexical representations but which is expected to be very useful in a wide range of NLP purposes. We show how this information can be used in dictionary sense disambiguation by creating links throughout the lexicon both on the paradigmatic and the syntagmatic axes.