Fast base NP chunking with decision trees - Experiments on different POS tag settings

Abstract

We introduce a chunking system based on probability distribution approximation and decision trees. The system was tested on the large standard evaluation set for base noun phrases. While the training time of the system is remarkable short (about 3 hours), results are comparable with the best systems reported so far. We trained our system with different settings of POS tags and show how much chunking results depend on POS tag accuracy. © Springer-Verlag Berlin Heidelberg 2003.