Hybrid NLG in a generic dialog system

Abstract

Natural Language Generation (NLG) systems are increasingly becoming available as "market-ready" products, mainly due to the now-removed boundary between shallow and deep generation and the emergence of hybrid systems as a de-facto standard. In this paper, we present HYPERBUG1, a novel approach towards hybrid NLG, coupling shallow and deep processing not only with respect to the resources used for parsing and generation, but also on the architectural level to increase the generative power of the shallow generation branch and the processing efficiency of the whole generation system. The architecture is discussed both in theory and in practice, using a comprehensive example spanning the complete output part of our dialog system. © Springer-Verlag 2004.