A Conversational-Mode Airline Information and Reservation System Using Speech Input and Output

Recently there has been a substantial research effort in the area of speech understanding. Although the ultimate purpose of this work is clearly that of enabling natural spoken language human/machine communication, most of the work has actually been in the nature of building systems which transcribe speech. The system described in this paper is capable of conducting a complete spoken dialog with its user. The essential system architecture is hierarchical with three levels. These are an acoustic word recognizer, a syntax analyzer, and a semantic processor. The semantic level controls an audio response system which provides the speaking function. There is, of course, a significant interaction of the levels with one another. The precursor to and incentive for this project was the flight information system built by Rosenberg and Itakura,1 which consists of only a word recognizer and a voice response unit but which nonetheless is capable of conducting a limited dialog. The first two levels of the 119