Natural Language Generation (NLG) technology has reached a level of maturity where applied systems exist in a range of specialised real-world domains (such as weather bulletins, software documentation, health and legal advice and stock market movements). However, developing such systems currently involves hand-crafting and special-purpose tuning by NLG experts which is non-portable, non-scaleable, time-consuming and expensive. Wider deployment of language generation requires more generally applicable and reusable NLG components based on wide-coverage grammars, but at present, effective techniques for such wide-coverage generation are not well understood. This three year project will investigate systematically the characteristics of wide-coverage generation and to develop reflective techniques for controlling it effectively. As well as furthering our understanding of wide-coverage generation, the project will deliver a substantial and novel resource to support future research in this area, and practical implementations of wide-coverage controllable generators.
Funding body: EPSRC
Duration: 2003-2006
This is a joint project involving University of Sussex and University of Brighton.
COGENT Team at Sussex: David Weir, John Carroll, Daniel Paiva and Eva Esteve Ferrer
COGENT Team at Brighton: Roger Evans, Kees Van Deemter and Anja Belz
We are collaborating informally with colleagues in the LOGON machine translation project in Norway.