Artificial Intelligence Review 8: 431-445, 1994-5. © 1995 KluwerAcademic Publishers. Printed in the Netherlands. The DenK-architecture: A Fundamental Approach to User-Interfaces R. M. C. AHN, R. J. BEUN, T. BORGHUIS, H. C. BUNT and C. W. A. M. VAN OVERVELD* Abstract. In this paper we present the basic principles underlying the DenK-system, a generic cooperative interface combining linguistic and visual interaction. The system integrates results from fundamental research in knowledge representation, communi- cation, natural language semantics and pragmatics, and object-oriented animation. Our design incorporates a cooperative and knowledgeable electronic assistant that communicates with a user in natural language, and an application domain, which is presented visually. The assistant, that we call the cooperator, has an information state that is represented in a rich form of Type Theory, a formalism that enables us to model the inherent cognitive dynamics of a dialogue participant. Pragmatic issues in man-machine interaction, concerning the use of natural language and knowledge in cooperative communication, are central to our approach. Key words: multimodal interaction, knowledge representation, natural language semantics, pragmatics, type theory, object oriented animation. 1. INTRODUCTION In 1989, the universities of Tilburg and Eindhoven initiated a joint research program that aims at the development of a multimodal cooperative interface for interactive knowledge acquisition and -analysis. Multimodality refers here to a combination of linguistic and visual interaction. This 8-year program is called 'Dialoogvoering en Kennisopbouw', abbreviated DenK, 1 which means 'Dialogue Management and Knowledge Acquisition'. The program combines fundamental research in knowledge representation, communication, natural language semantics, pragmatics and object-oriented animation to develop generic user interface techniques. These techniques are applied in the prototypical DenK-system. The design of this system reflects a cooperative situation where two participants can exchange information about a shared application domain and where both participants can point at, observe and manipulate objects within this domain (cf., Grosz 1978). The point of departure is that, from a user's point of view, a computer appli- cation should present itself as a combination of an 'electronic assistant' and a tangible model of the task domain. The electronic assistant should interact in an intelligent and cooperative way with the user, using a combination of linguistic and visual modalities, and acting on the user's intentions as understood in the context of the interaction, 267