In this paper, we show an approach to extracting \ndifferent types of constraint rules \nfrom a dependency treebank. Also, we \nshow an approach to integrating these constraint \nrules into a dependency data-driven \nparser, where these constraint rules inform \nparsing decisions in specific situations \nwhere a set of parsing rule (which is \ninduced from a classifier) may recommend \nseveral recommendations to the parser. \nOur experiments have shown that parsing \naccuracy could be improved by using different \nsets of constraint rules in combination \nwith a set of parsing rules. Our parser \nis based on the arc-standard algorithm of \nMaltParser but with a number of extensions, \nwhich we will discuss in some detail.