Text Planning of Air Quality Information

Abstract

Two of the premises of the Multimodal Air Quality Information System known as MARQUIS are that (1) air quality information is dynamic (i.e., it might change every hour) and different data variations should be expressed differently, (2) people have different expertises, needs and interests about air quality and this must be reflected in the content and style in which this information is conveyed to them. The text planning module in MARQUIS achieves precisely that: it selects the content that is of relevance to a specific user from the mass of air quality information produced by the interpretation module, and arranges this information into a coherent discourse that takes into account the dynamic content and the profile of the user in the required platform. In this paper, we describe how we addressed these issues of dynamic content selection (section 2) and user modelling (section 3) in the text planning system of the MARQUIS project. These play an active role in the production of the ouput text plan (section 4) as they contrain the order of propositions and the types of discourse relations that occur between the propositions (section 5). We describe the architecture of the text planner (section 6) before giving some conclusions (section 7).