The influence of time on error-detection

Abstract

Secretaries are needed to type out the reports dictated by medical specialists. Even with a large transcription department it takes more than a day before a dictated report is ready to be sent to the people concerned. The use of automatic speech recognition (ASR) can shorten this time interval. A disadvantage is that ASR systems still make errors. As reports must be error-free a lot depends on the pathologists' correction capabilities. The correction procedure that pathologists currently use may not be adequate for correcting ASR errors. To be able to correct the reports pathologists must be able to recollect the structure and contents of their dictation. Their recollection is likely to be better with less time between dictation and correction. In this paper correction performance on the same day as dictation is compared with correction the day after. To be able to make this comparison, errors were artificially introduced into the pathologists' reports. No difference between the conditions was observed, but it was found that pathologists had great difficulty correcting the reports. It is concluded that their current correction procedure is inadequate to correct ASR errors and may be inadequate altogether.