Robert Dale, Maria Milosavljevic, and Jon Oberlander
A very common metaphor used when discussing the World Wide Web, and hypertextual systems in general, is that of navigation within a space. This is not the only way to think about hypertext, however: an alternative, and potentially richer, metaphor is to view a session with any such system as a dialogue, with the user and the system taking alternate turns in the conversation. This acknowledges the interactive nature of the experience, and may open up new ways of thinking about how the medium can be used. Unfortunately, existing hypertext systems are not up to the task. A true dialogue requires a conversational participant who is able to take account of the content of the dialogue so far, and to respond appropriately in context; existing hypertext systems are, by and large, providers of pre-written, canned material. By using natural language generation techniques, however, we can tailor the system’s contributions to take account of the ongoing discourse context. In this paper, we look at how natural language generation techniques can be used to enhance the coherence of the system’s turns in the conversation. We demonstrate the ideas by reference to two Web-based language generation systems we have constructed, the PEBA-[I system and the [LP.X system.