Combining Discourse Representation Theory with FrameNet

Discourse Representation Theory (DRT) by and large ignores the issue of thematic roles. We have investigated the possibilities of combining DRT with thematic roles such as those from FrameNet, both from a theoretical and practical perspective. We show that a neo-Davidsonian version of DRT matches best with FrameNet's intentions, but also demonstrate that not all FrameNet roles can be meaningfullly expressed in the representation language of DRT, in particular those introduced by adverbs. Using a wide-coverage implementation of DRT, we present an algorithm for learning FrameNet roles and adding these to the meaning representations produced by a state-of-the-art wide coverage parser.