This paper reports the results of the development, deployment and test
ing of a large spoken-language dialogue application for use by the gen
eral public. We built an automated spoken questionnaire for the US Bur
eau of the Census. In the project's first phase, the basic recognizers
and dialogue system were developed using 4000 calls. In the second ph
ase, the system was adapted to meet Census Bureau requirements and dep
loyed in the Bureau's 1995 national test of new technologies. In the t
hird phase, we refined the system and showed empirically that an autom
ated spoken questionnaire could successfully collect and recognize cen
sus data, and that subjects preferred the spoken system to written que
stionnaires. Our large data collection effort and two subsequent field
tests showed that, when questions are asked correctly, the answers co
ntain information within the desired response categories about 99% of
the time.