Subproject 2: Frame validation and cross-lingual frame projection

Even though frames themselves are considered universal, the way they are expressed differs from language to language. The lexical realization of a frame structure is crucial, as it provides the surface access to the frame itself, both in the acquisition and in the interpretation. This subproject will investigate the issue cross-linguistically, starting from already developed English FrameNet resources and mapping these to (primarily) Dutch and (secondarily) Italian. The two objectives of this subproject are (1) an automatic statistical Frame-labeller for the Dutch language, and (2) an annotated parallel corpus of sentences with frame semantics for the three languages. Both objectives depend on each other: in order to develop the frame-labeler (which will be based on supervised learning), we need texts annotated with frames. We will use bootstrapping and annotation projection as a general method.