Statistical Parsing of Dutch using Maximum Entropy Models with Feature Merging