To Örebro University

kLog is a framework for kernel-basedlearning that has already proven success-ful in solving a number of relational tasksin natural language processing. In this pa-per, we presentkLogNLP, a natural lan-guage processing module for kLog. Thismodule enriches kLog with NLP-specificpreprocessors, enabling the use of exist-ing libraries and toolkits within an elegantand powerful declarative machine learn-ing framework. The resulting relationalmodel of the domain can be extended byspecifying additional relational features ina declarative way using a logic program-ming language. This declarative approachoffers a flexible way of experimentationand a way to insert domain knowledge.