Speech Understanding Research

Abstract

Abstract : The following summary provides a perspective on the distinctive characteristics of the Stanford Research Institute contributions to the current system. The system control, embedded in the parser, focuses the operation of the entire system to minimize both storage requirements and the time spent on incorrect interpretations. A language definition system provides a means for integrating the various sources of knowledge in the system. The language definition itself, based on studies of protocols gathered from actual performances in task-oriented dialogs, includes information from acoustics, phonetics, phonology, prosodics, syntax, semantics, pragmatics, and discourse. A new semantic network representation, which partitions the net into spaces, has proved particularly well suited for working with the two task domains. Discourse procedures, building on the semantics, establish a discourse history so that information from previous utterances (and, ultimately, from the task environment) can be used in the analysis of the current utternce. Appendix A contains a listing of the language currently defined in the speech understanding system. Appendix B lists the reports and publications issued by the project staff. Additional keywords: computer applications.