Effects of complexity and type of information in the detection of different types of error 

P.M. Schellekens
In the future it is expected that more and more Automatic Speech Recognition Systems (ASR-systems) will be used. (ASR means that dictated words will be identified and transposed into text by a speech-recognition system.) While the systems gather strength, the suitability of ASR will increase. One of the possible applications can be found in the medical field of diagnosis where work efficiency is a primary issue.

Nowadays pathologists examine tissues under a microscope and dictate the results on tape. This tape is then transcribed by a secretary and the resulting manuscript is send back to the pathologist for correction. In the future it is likely that these secretaries may be replaced by ASR-systems. The Institute for Perception Research (IPO) is investigating the demands of a system which enables the pathologists to print their dictated reports on line. One of the questions brought up by this study was: What determines the vigilance of the pathologists for errors made by the ASR-system?'.

The report under discussion describes an experiment in which the task of the pathologists was simulated. Subjects had to classify pictures while dictating reports on tape. The next day they were asked to correct their dictated reports on errors while the subjects received additional information. The concerned variables were:

  1. Complexity, this comprised the level of difficulty of the classifying task.
  2. Information, this comprised the modality of the additional information provided during correction.
  3. Error, this comprised the four types of error which may arise during speech-to-text transposition.
The study was mainly used to improve and adjust the model which describes the stages involved with memorising reports by pathologists.

05-04-2000

  Info: