BBN HARC and DELPHI Results on the ATIS Benchmarks - February 1991

This paper presents the test results of running BBN's HARC spoken language system and DELPHI natural language understanding system on the ATIS benchmarks. We give a brief system overview, and review the major changes that have been made in Delphi since the last DARPA SLS workshop. We will briefly discuss the development and training process, and then present our test results and an analysis of their meaning. S Y S T E M O V E R V I E W Delphi is BBN's research NL system, which is based on a unification grarnrnar and which incorporates semantics into the unification framework. Delphi is the NL component of the BBN HARC (Hear and Respond to Continuous Speech) system; integrated with the BYBLOS speech recognition system using an N-best architecture [1,2]. Figure 1 shows the relationships among the components of HARC, and their inputs and outputs.