We propose a method for rapid development of dialogue systems where a Grammatical Framework (GF) grammar is compiled into a complete VoiceXML application. This makes dialogue systems easy to develop, maintain, localize, and port to other platforms, and can improve the linguistic quality of generated system output. We have developed compilers which produce VoiceXML dialogue managers and EC-MAScript linearization code from GF grammars. Along with the existing GF speech recognition grammar compiler, this makes it possible to produce a complete mixedinitiative information-seeking dialogue system from a single GF grammar. 1

Grammatical Framework (GF) is a meta-language for multilingual linguistic descriptions, which can be used to build rule-based interlingua MT applications in natural sublanguage domains. The GF open-source package contains linguistic and computational resources to facilitate language engineering including: a resource grammar library for ten languages, a user interface for multilingual authoring and a grammar development environment. 1

Abstract. We describe how to give a full specification of an ISU-based dialogue system as a grammar. For this we use Grammatical Framework (GF), which separates grammars into abstract and concrete syntax. All components necessary for a working GoDiS dialogue system are specified in the abstract syntax, while the linguistic details are defined in the concrete syntax. Since GF is a multilingual grammar formalism, it is straightforward to extend the dialogue system to several languages. Furthermore, the GF Resource Grammar Library can be used to write a single concrete instance covering 13 different languages. 1

A Platform-Independent Model (PIM) is supposed to capture the requirements specified in the Computational Independent Model (CIM). It can be hard to validate that this is the case since the stakeholders might lack the necessary training to access the information of the software models in the PIM. In contrast, a description of the PIM in natural language will enable all stakeholders to be included in the validation. We have conducted a case study to investigate the possibilities to generate natural language text from Executable and Translatable UML. In our case study we have considered a static part of the PIM; the structure of the class diagram. The transformation was done in two steps. In the first step, the class diagram was transformed into an intermediate linguistic model using Grammatical Framework. In the second step, the linguistic model is transformed into natural language text. The PIM was enhanced in such a way that the generated texts can both paraphrase the original software models as well as include the underlying motivations behind the design decisions. Categories and Subject Descriptors