When humans use language, they show an essential responsiveness to their hearers. When language is automatically generated, it is similarly necessary to ensure that language is appropriate for its intended audience. Much previous research in computational linguistics has focused on building user models and selecting information from a knowledge base to present to the user. It is important, however, that the phrasing of a text be also tailored to the hearer - otherwise it may be just as ineffective as texts which rely on knowledge the hearer does not have. The realization that situations systematically affect language can greatly help the tailoring process: given appropriate classifications of situation-types, the consequences of these situations on language can be specified, and suitable language can be produced. In a pilot study that involved generating explanations for an expert system tailored to three user types, previous work has shown that is is possible to specify the type of language required in a given situation and to build a text planning system that uses that specification to control phrasing of its text. This research will extend that work. The goal is to create a framework within which it is possible to gain systematic control over phrasing by designing methodologies and mechanisms by which the systematic relationship between situations and language can be found, represented, and used to tailor generated texts.
