A demonstration of a natural-language pedestrian routing system

This demonstration highlights some features of a dialogue system for pedestrian routing (Boye et al. 2012) that uses natural-language routing instructions to help pedestrians navigate and explore the city in real-time. The instructions are given as the user is moving: When the user reaches a node pi in the planned route, he is informed how to reach the next node pi+1. The user can ask for (more) instructions if he is unsure where to go next, and reply to the system’s questions (by “yes” or “no”). The system keeps track of the user’s position by GPS relayed from the user’s smartphone. Obviously, it is increasingly interesting to create and study such applications as more and more people have smartphones equipped with a GPS receiver. In order to make the system scalable and transferrable across cities, the system uses only open, crowd-sourced geographical data from OpenStreetMap (Haklay 2008). Such data is admittedly more error-prone and not as consistently organised compared to data released from a governmental surveying organization, but on the other hand, national surveying data tends to be expensive, hard to access, and vary in format from nation to nation. Moreover, the possibility to port the system to different cities is important from a research point of view. Different cities have different characteristics (narrow streets or big boulevards, a grid-like structure or a medieval city plan, etc), and these features will influence the success of different system strategies for producing route instructions. A key problem for routing systems is the generation of natural language references to objects in the city. Such references form the link between the algebraic and geometric model of the domain on the one hand, and the communication with the user on the other. The strategy employed by our system is to, as far as possible, describe the route using landmarks, by which we understand distinctive objects in the city environment. Street names in instructions are avoided unless there