Kybernetika 40 no. 3, 349-363, 2004

Communication with www in Czech

Lukáš Svoboda and Luboš Popelínský

Abstract:

This paper describes {\em UIO}, a multi-domain question-answering system for the Czech language that looks for answers on the web. {\em UIO} exploits two fields, namely natural language interface to databases and question answering. In its current version, {\em UIO} can be used for asking questions about train and coach timetables, cinema and theatre performances, about currency exchange rates, name-days and on the Diderot Encyclopaedia. Much effort have been made into making addition of a new domain very easy. No limits concerning words or the form of a question need to be set in {\em UIO}. Users can ask syntactically correct as well as incorrect questions, or use keywords. A Czech morphological analyser and a bottom-up chart parser are employed for analysis of the question. The database of multiword expressions is automatically updated when a new item has been found on the web. For all domains {\em UIO} has an accuracy rate about 80 \