The ZAS's main contribution to this project will be based on results recently obtained at the research centre in the field of formal pragmatics of natural language. By incorporating these results, the project aims at enhancing communication with NPCs with more naturalness and more intuitivity for the player. We particularly focus on the modelling of pragmatic phenomena such as the generation of optimal answers and the interpretation and generation of implicatures. This is not only of interest to the primary target application but also to the development of more intuitive interfaces for human-machine communication.
The use of a hybrid architecture, which combines deep syntactic-semantic analysis with robust shallow-processing methods, guarantees a functional base system at any time, which nevertheless can be incrementally enhanced at many levels with increasingly abstract and, thus, more intelligent components. The combination of linguistic processing with domain and world knowledge will allow to adapt the linguistic capabilities of an NPC to specific virtual contexts depending on the role played by it.
Furthermore, exemplary test NPCs should be made available for linguistic research purposes, in order to serve as experimental platforms for the empirical verification and refinement of the theoretical pragmatic models.
By partnerships with companies of the Berlin computer game industry, it is guaranteed that the solutions, evaluation criteria and technological results comply with the requirements and demands of commercial application. In this way, Metaversum GmbH, for example, whose virtual world Twinity imposes particularly high requirements in terms of naturalness due to its closeness to reality, supports the project by licensing real character models.
Prof. Dr. Manfred Krifka