Generating Tailored, Comparative Descriptions in Spoken Dialogue
Johanna Moore

In this talk, I will describe an approach to automatically generating user-tailored spoken descriptions with appropriate intonation. The approach brings together multi-attribute decision models for user modelling, strategic content planning, and a realisation component, OpenCCG, that uses information structure to produce texts with markup for controlling prosodic features during synthesis. This talk emphasizes how the user model influences virtually every aspect of the generation process, including selection of the most important subset of available options to mention and the attributes that are most relevant to choosing between them; organization of the selected content; expression of descriptions of selected options and attributes; and selection of intonational features.