Inferring implicatures and discourse relations from frame information

Two types of implicit meaning aspects play an important role in discourses comprising more than one utterance: implicatures that arise when the linguistic input is minimized, and inferences that establish connections between utterances in order to form a coherent discourse. This paper discusses how these inferences can be modelled in terms of integrating a formal theory of discourse structure and interpretation, Segmented Discourse Representation Theory, with FrameNet, a cognitive network of stereotypical situations. This combination allows us to draw a variety of implicatures in a setting that takes possibly complex discourse structures explicitly into account. Underspecified semantic forms derived from the linguistic input are enriched by parameters given by FrameNet information. Implicit information is represented as open parameters which are to be filled either by default values or by subsequent information in the discourse. The approach covers phenomena like free enrichment and bridging, and also helps inferring implicit discourse relations.