Helping motorists from police reports - Poetic

A typical example of a symbolic AI application. ``Portable Extendable Traffic Information Collator''. A natural-language system which reads police incident logs and selects out information about those traffic problems which might concern other motorists. From these, builds up a coherent description of each incident, and then works out how motorists in the area might avoid the problem. Converts the advice into English, and then decides where best to broadcast each message. See POETIC: A System for gathering and distributing traffic information by R Gaizauskas, R Evans, L J Cahill, J Richardson, J Walker, photocopy G87 in the AI box: these include examples of the police logs that serve as input, and the advice generated as output, and are easy to read even if you are new to AI.

Some points about Poetic:

• One thing it must do is to build up a world model: a coherent description of the world, derived from its sensory input. It's been a common assumption in AI (now being challenged, see footnotes in my first tutorial handout and the PopBeast stuff with it) that this is a necessary part of any intelligent organism, and that animals and humans will embody such a model. Like many other symbolic AI programs, Poetic's world-model is in logic, in this case via the Prolog programming language.

• This may involve cross-referencing different log entries to disambiguate references to objects. For example ``8.05: lorry crash on A413'' and ``9.17: road is still blocked''. Are these the same road? Illustrates a general problem: perceptions are usually incomplete. If an agent's world model is going to represent individual objects, the agent will have to match up different partial descriptions to determine whether they refer to the same object. These descriptions may come from different sensors at the same time, or the same sensor at different times.

• The natural-language understander must cope with ungrammatical input and jargon. Police logs are not standard English! Also, not all the information in the logs is about traffic problems, so Poetic must analyse the entries before it knows whether to ignore their content.