This paper describes a technique for both constituent and dependency parsing. Parsing proceeds by adding functional labels to the output of a constituent parser trained on the French Treebank in order to further extract typed dependencies. On the one hand we specify on formal and linguistic grounds the nature of the dependencies to output as well as the conversion algorithm from the French Treebank to this dependency representation. On the other hand, we describe a class of algorithms that allows to perform the automatic labeling of the functions from the output of a constituent based parser. We specifically focus on discriminative learning methods for functional labelling.