A history-taking system that uses continuous speech recognition

Abstract

Q-MED is an automated history-taking system that uses speaker-independent continuous speech as its main interface modality. Q-MED is designed to allow a patient to enter her basic symptoms by engaging in a dialog with the program. Error-recovery mechanisms help to eliminate findings resulting from misrecognitions or incorrect parses. An evaluation of the natural language parser that Q-MED uses to map user utterances to findings showed an overall semantic accuracy of 87 percent; Q-MED asks more specific questions to capture findings that were not volunteered, or that were unable to be parsed in their initial, open-ended form.