Alignment-Based Learning versus Data-Oriented Parsing