Robust Pragmatics for Narrative Text, Research Project Grant, funded by the EPSRC, 01/2002-03/2005.

Principal Investigators: Alex Lascarides, Mirella Lapata
Research Fellow: Caroline Sporleder

In this project we will develop a hybrid approach for automatically identifying the temporal structure of discourse by combining data-intensive methods and constraints on the well-formedeness of discourse structures. We will model temporal progression by exploiting a hybrid system, interfacing a probabilistic component with a symbolic one. We will use probabilistic modelling to combine multiple sources of linguistic knowledge (i.e., features) extracted automatically from large corpora to infer rhetorical relations. The symbolic component will specify how the rhetorical relations have specific semantic effects on the text, particularly its temporal content. Such an approach can deal with domain independent narrative text, it is robust, and relatively inexpensive as it assumes that crucial pragmatic features can be discovered by analysing the linguistic environment, as approximated by large corpora.