Exploiting pauses in continuous speech recognition

A primary application of the study of hesitation phenomena lies in improving the performance of automatic recognizers, given an input of spontaneous speech (e.g., verbal conversions with computer databases). Speech researchers have often expressed interest in exploiting the intonation of spoken utterances in the recognition process, but have been deterred by the complex nature of how intonation (including pauses) relates to the text of an utterance. Even straightforward phenomena such as unfilled pauses (i.e., silence periods-which are generally easy to identify, if long enough) are not reliable indicators to the syntactic or semantic sentence structure of an utterance