In this article, we describe an automatic system for train timetable i
nformation over the telephone that provides accurate connections betwe
en 1200 German cities. The caller can talk to it in unrestricted, natu
ral, and fluent speech, very much like he or she would communicate wit
h a human operator, and is not given any instructions in advance. The
system's four main components, speech recognition, speech understandin
g, dialogue control, and speech output, are separated into independent
modules that are executed sequentially. Word graphs form the interfac
e between recognition and understanding; an attributed stochastic cont
ext-free grammar is then used to determine the meaning of a spoken sen
tence. In an ongoing field trial, this system has been made available
to the general public, both to gather speech data and to evaluate its
performance. This field test was organized as a bootstrapping process:
initially, the system was trained with just the developers' voices, t
hen the telephone number was passed around within the department, the
company and, finally, the outside world. After each step, the newly co
llected material was used for retraining, as well as for general impro
vements.