Does natural selection apply to natural language processing? an experiment for multiword unit extraction

In this paper, we focus on the suitability of natural selection for the extraction of Multiword Units (i.e. complex lexical units such as compound nouns, idiomatic expressions or phrase templates). For that purpose, a fitness function is defined whose maximization serves as a basis for the identification of pertinent word N-grams together with a similarity measure. In order to propose a suitable platform for evaluation, a software application called GALEMU (Genetic ALgorithm for the Extraction of Multiword Units) has been implemented. Finally, we will provide an experiment realized over an unnnotated text corpus extracted from the database collection of the European Commission that evidences results with high precision rate