We first provide some experimental data concerning language shifts. The experimental setting used consists of two agents, an instructor and a constructor. The instructor describes a blocks' world to the constructor, who builds it up according to the instructor's directives (ch. 1). In ch. 4 we describe the architecture of our system SPEX (speech exchange system), which models perception, focussing and intentions of situated agents as well as their production and reception of directive speech acts. Special attention is devoted to language shifts and indirectness phenomena. The solutions incorporated into the simulation engine SPEX are evaluated against suggestions found in the literature. The topics touched upon in this context are agents' ontologies, perception, intention, focus, speech act processing, coordination of mental states and actions and language shifts (ch. 2). Considerable care is given to the description of the focus model integrated in SPEX (ch. 3). Chapter 4 presents the interacting modules of SPEX' agents' processes down to the level of detail needed in order to grasp the working of the simulation engine. Here focus is laid on the information flow within an agent and the transfer of information established between agents. We also explain which mechanisms simulate the agents' production and reception of language shifts (ch. 5). The generation of e.g. metonymical expressions of various sorts (rectangle for cuboid or square for cube) is based on special rules using salience and functional relevance of objects as well as the perspective of the observing agent. An agent's understanding of metonymies is modelled by Gricean conversational implicatures. In a ``simulation session'' the exchange of speech acts is described in some detail including the representations needed on various levels of cognition (ch. 6). In ch. 7, the focus hypotheses as specified by the simulation engine are confronted with experimental results based on an eye-tracker study using the same empirical setting. The study showed that there is a lag between intending, focussing and planning of speech acts on the one hand and speech production on the other hand. It also revealed that agents coordinate mental activities like focussing objects. Finally, we discuss how these findings can be incorporated into the simulator engine.