Model driven development approach to natural language generation systems

Abstract

Natural Language Generation systems have traditionally been built using ad-hoc software engineering practices with no explicit development process and no standard software architecture. This situation has drastically limited professional use of NLG technology. New approaches to NLG application development that considers domain-independence, languages and standards of modern software engineering, could enhance its practical use. This work proposes an NLG reference architecture that leverages the most advanced open standards in software architecture, modeling languages and processing tools. In particular, it is shown how dialogue-based voice-driven NLG applications can be built from the up-to-date Model-driven Architecture (MDA) approach. The implementation of a voice-driven movie recommendation system demonstrates the feasibility of the proposal.