Parsing and collocations

Abstract

Proper treatment of collocations constitutes a serious chal-lenge for NLP systems in general. This paper describes how Fips, a "Principle and Parameters" grammar-based parser developed at LATL handles multi-word expressions. In order to get more precise and more reliable collocation data, the Fips parser is used to extract collocations from large text corpora. It will be shown that collocational information can help ranking alternative analyses computed by the parser, in order to improve the quality of its results.