PUB - Publikationen an der Universität Bielefeld

The goal of this thesis is to simulate cooperation in human-computer communication to model the communicative interaction process of agents in natural dialogs in order to provide advanced human-computer interaction in that coherence is maintained between contributions of both agents, i.e. the human user and the computer. This thesis contributes to certain aspects of understanding and generation and their interaction in the German language. In spontaneous dialogs agents cooperate by the production of utterances. Accordingly, not only independent utterances are produced but an utterance which is produced by one agent can be continued by the other one based on the structural properties provided so far by the initial utterance. As this thesis concentrates on the underlying syntactic structures of these interactive processes which are carried out cooperatively by both agents of a dialog the aspects of human communication under consideration will be called "cooperative syntax production". In accordance with the interactive character of these processes a computational system for modeling the interaction between agents of a dialog must integrate both aspects of natural language: understandig and generation. Consequently, this system must coordinate the understanding and generation process in a cooperative way so that it is possible to switch flexibly between both processes back and force. To perform this interaction task an understanding and generation component has to be developed which meets the fundamental condition of cooperative syntax production or is at least able to demonstrate certain aspects of the underlying skills. As this thesis focuses on coordinate syntax production the understanding component will be restricted mostly to syntax. But for the reinterpretation process and for generating an adequate continuation of an initial utterance the model must have access to some semantic or pragmatic knowledge. This knowledge is given by visual context information which enables the system to derive and generate context sensitive interpretations of a given situation. Summarizing, this thesis presents a model for simulating cooperative syntax production. The model is based on an iterative processing strategy which is interactive in means of flexible changing from parsing to generation and vice versa. Besides, the integration of visual context information into the model establishes a robust and goal-directed processing strategy. Moreover, the presented model is robust to certain phenomena of spontaneous speech and to speech recognition errors.