This abstract outlines work in progress at the University of Durham concerning natural language processing. As a result of this work, a general theory has been developed which offers a means of modelling natural English dialogue by computer, and can be used for both the analysis and generation of dialogue. The implementation of this theory is currently in progress (commenced in early 1993) and forms a dialogue module which uses input from a large NLP system developed at Durham. The theory categorises natural dialogue into template-like schemas (called Dialogue Structure Models) which contain basic information about different types of dialogue, (e.g. lecture, interview etc.). This information provides the system with knowledge about the structure of the dialogue and how it comprises of its constituent parts, which we call Dialogue Elements (DE). The DEs are factors which influence and control the dialogue structure. For example, the Dominance DE informs the system as to which participant(s) has control within the dialogue. There are fourteen DEs within the theory, each carrying different information. Their combinations vary for each dialogue participant, depending upon the role being played. The DEs inform the system as to what is "the norm" within a given DSM. However, the dialogue system must be capable of utilising this information, and this requirement is fulfilled by attaching a set of "constraints" to each Dialogue Element. Each constraint is a possible action which the system may take, (e.g. if a participant has dominance, he/she may ask a question, change the topic etc.). The way in which the system selects the appropriate action to take depends upon a number of influencial factors, including the system's "personality", previous knowledge held about the user, and particular details of that individual dialogue etc. Once the system has selected the appropriate constraint, this information is passed to the generation module, translated into English and displayed upon the screen for the user to read and respond to. The process is then iterated, developing a dialogue.
Keywords: Dialogue Structure, Natural Language Processing, Constraints, Planbox