DARPA FEBRUARY 1992 ATIS BENCHMARK TEST RESULTS

[2]. series were answers. were swers (Ver- SLS (e.g., "over- 3 • Since were 15 1991 20899 false bles, were 1991 • 1991 were A DahIgren, a released. (2) G. [1], were 3.3). queries collec- • 1 were fixes, D clar- Class Speech Garofolo, a (i.e., • C. 1990 files success- • a was 2 • a (close- (3) • (1) collection 1991 five were 225, • DARPA FEBRUARY 1992 ATIS BENCHMARK TEST RESULTS David S. Pallett, Nancy L. Jonathan Fiscus, William M. Fisher, John S. Brett Tjaden National Institute of Standards and Technology Building Room A216 Gaithersburg, MD INTRODUCTION ify terminology, revise headings and restructure ta- improve representation of fare structures, bug This paper documents the third in of Bench- etc) made to the relational air-travel- mark Tests for the DARPA Air Travel Information Sys- information database. The MADCOW data tem (ATIS) common task domain. The first results in tion effort, and systems developed with this data, this series reported at the June 1990 and made use of this revised relational database Natural Language Workshop and the second at the sion February Speech and Natural Language Workshop The MADCOW data collection effort provided data The February 1992 Benchmark Tests include: from sites (AT&T, BBN, CMU, MIT/LCS, and ATIS domain spontaneous speech recognition system SRI), rather than the single ATIS data tests, ATIS natural language understanding tests, site (TI) used ...