This article presents a method for automatic assignment of syntactic dependency relations to the corpus of American Norwegian speech (CANS). Different machine learning techniques and corpora are used. Finally, an accuracy measure is computed and compared with a relatively new treebank for spoken Norwegian.