The application of Genetic Algorithms in Part-of-Speech tagging for Spoken Dutch
Wouter Joosse · 2006
way to train automatic Part-of-Speech Taggers is using Brill's Transformation-based Learning approach. A number of Part-of-Speech taggers for various languages have been developed with this method, and these taggers usually have high accuracies. The standard Transformation-based Learning method has a disadvantage though: it has a long learning time, due to a rather costly search process, looking for the best set of transformation rules. This paper reports on studying ways to improve the efficiency of selecting the best set of rules using a Genetic Algorithm. It is shown in an experimental way how the accuracy of the tagger achieved by a Genetic Algorithm depends on the size of the population, the mutation chance and on the selection method that is used.