Maximising the Coherence of Descriptions via Comparison

In this thesis, we explore the use of comparison in descriptions of entities. We postulate that entity description is largely a process of comparison with and distinction from familiar and similar entities in order to maximise the hearer's conceptual coherence of that entity, and to prevent the hearer from forming misconceptions about that entity. The process of producing a description of an entity requires the use of knowledge about the domain and situational context in order to ensure that the description is understood by the hearer. In this thesis, we analyse the use of comparison in descriptions of entities, and build computational models which support the incorporation of meaningful comparisons in the generation of descriptions. We employ information about the hearer's existing knowledge of the world in order to build on her understanding rather than to simply use her knowledge as a check, not repeating what she already knows. This thesis makes the following key contributions: Classi cation of the di erent types of comparison which can be used in descriptions of entities, and their roles in the description process. Examination of the broader role of comparison in the automatic generation of discourse, and algorithms for automatically generating meaningful comparisons in descriptions of entities. Analysis of the requirements for a similarity metric which is useful for producing textual comparisons of entities, and a similarity metric which takes these into account. xii