Chris Mellish, David Allport, Anthony F. Hartley, Roger Evans, Lynne J. Cahill, Robert Gaizauskas, John Walker
The Traffic Information Collator (TIC) is a software system that extracts traffic information from free text in police incident logs and initiates (simulated) broadcasts of traffic bulletins to motorists as appropriate. This paper describes the initial message analysis module of the system: a robust, selective chart parser operating on a special-purpose grammar which explicitly encodes the kinds of reduced syntactic forms and nominalisations that occur within the domain of traffic incident messages. The prototype system takes as input actual incident logs provided by the UK Sussex police force. These are telegrammatic in style, with extensive use of jargon and abbreviations, and they are constructed and need to be reacted to in real time. Many of the messages in the incident logs are unrelated to traffic, and even those which are relevant often contain arbitrary fragments of text of no interest to the system. Thus the task of the initial message analysis is to extract useful information robustly and quickly, while ignoring the rest.
This paper is not available online