Representing and using protosemantic information in generating bus route descriptions

Abstract

In this paper we examine the domain of route communication, and propose a representation for the bus route map of a city and an effective route finding procedure for use in the context of generating bus route descriptions in natural language. We present a concise analysis of bus route descriptions from the perspective of their information structure. We have noted that route descriptions involve information of different degrees of granularity, and that in the case of bus route descriptions, the activation of coarse-grained information can be separated from that of finer-grained information. Accordingly, we represent the map mainly as a collection of routes. We use simple set operations at the core of our route finding procedure. We have adapted these operations to extract relevant annotations from the knowledge base for generating text plans. We briefly discuss how the spatial problem of route finding can be treated as a search problem in the context of giving bus route directions. Finally, we consider bus route description generation as a candidate problem domain for knowledge based systems. We have implemented our knowledge base design and the route finding procedure in a prototype system that finds bus route directions in Vancouver, Canada.