In this article we want to demonstrate that annotation of multiword expressions in the Prague Dependency Treebank is a well de- fined task, that it is useful as well as feasible, and that we can achieve good consistency of such annotations in terms of inter-annotator agreement. We show a way to measure agree- ment for this type of annotation. We also ar- gue that some automatic pre-annotation is possible and it does not damage the results. 1 Motivation Various projects involving lexico-semantic annota- tion have been ongoing for many years. Among those there are the projects of word sense annotation, usu- ally for creating training data for word sense disam- biguation. However majority of these projects have only annotated very limited number of word senses (cf. Kilgarriff (1998)). Even am...

Click Here to View the Paper In XML Format