Extracting Dependency Frames from Existing Lexical Resources

One method for evaluating a wide-coverage parser involves measuring how accurately it identifies dependency relations. The construction of a grammar which outputs dependency relations requires a lexicon with detailed information on subcategorization and dependency. We discuss how such a lexicon can be constructed automatically for a wide-coverage lexicalist grammar for Dutch by extracting the relevant information from existing lexical resources. We compare the coverage of the two sources relative to each other, and the coverage of the resulting lexicon with respect to a dependency treebank.