BusTUC - A natural language bus route oracle

BusTUC - A natural language bus route oracle Tore Amble Dept. of computer and information science University of Trondheim Norway, N-7491 amble@idi, ntnu. no Abstract The paper describes a natural language based expert system route advisor for the public bus transport in Trondheim, Norway. The system is available on the Internet,and has been intstalled at the bus com- pany's web server since the beginning of 1999. The system is bilingual, relying on an internal language independent logic representation. 1 Introduction A natural language interface to a computer database provides users with the capability of obtaining in- formation stored in the database by querying the system in a natural language (NL). With a natural language as a means of communication with a com- puter system, the users can make a question or a statement in the way they normally think about the information being discussed, freeing them from hav- ing to know how the computer stores or processes the information. The present implementation represents a a major effort in bringing natural language into practical use. A system is developed that can answer queries about bus routes, stated as natural language texts, and made public through the Internet World Wide Web ( http : //www. idi. ntnu. no/bustuc/). Trondheim is a small city with a university and 140000 inhabitants. Its central bus systems has 42 bus lines, serving 590 stations, with 1900 depar- tures per day (in average). That gives approximately 60000 scheduled bus station passings per day, which is somehow represented in the route data base. The starting point is to automate the function of a route information agent. The following example of a system response is using an actual request over telephone to the local route information company: Hi, I live in Nidarvoll and tonight i must reach a train to Oslo at 6 oclock. and a typical answer would follow quickly: Bus number 54 passes by Nidarvoll skole at 1710 and arrives at Trondheim Railway Station at 1725. In between the question and the answer is a pro- cess of lexical analysis, syntax analysis, semantic analysis, pragmatic reasoning and database query processing. One could argue that the information content could be solved by an interrogation, whereby the customer is asked to produce 4 items: station of departure, station of arrival, earliest departure timeand/or latest arrival time. It is a myth that natural language is a better way of communication because it is "natural language". The challenge is to prove by demonstration that an NL system can be made that will be preferred to the interrogative mode. To do that, the system has to be correct, user friendly and almost complete within the actual domain. 2 Previous Efforts, CHAT-80, PRAT-89 and HSQL The system, called BusTUC is built upon the clas- sical system CHAT-80 (Warren and Pereira, 1982). CHAT-80 was a state of the art natural language sys- tem that was impressive on its own merits, but also established Prolog as a viable and competitive lan- guage for Artificial Intelligence …

Content truncated to 500 words