Human performance modeling in temporary segmentation Chinese character handwriting recognizers

  • Authors:

  • Changxu Wu;Kan Zhang;Yongge Hu

  • Affiliations:

  • Engineering Psychology and Human Factors Laboratory, Institute of Psychology, Chinese Academy of Sciences, Beijing, PR China and Industrial and Operations Engineering Department, University of Mic ...;Engineering Psychology and Human Factors Laboratory, Institute of Psychology, Chinese Academy of Sciences, Beijing, PR China;Intel Research China Center, Beijing, PR China

  • Venue:
  • International Journal of Human-Computer Studies
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Human performance in Chinese character handwriting recognizers is critical to the satisfaction and acceptance of their users. Based on Teal's [CHI'92 (1992) p. 295] interactive model, a static model describing the independent factors in determining the task completion time was set up with a simple mathematical inference; in addition, a dynamic model describing these factors' direct and indirect causal relationship was established by the path analytic method. Results in Experiment 1 indicated that both the static model and the dynamic model could fit observed task completion time satisfactorily with minor modifications. In addition, with users' average writing time around 1500 ms for each frequently used character, it was found that the user's performance was impaired significantly when segmentation time was longer than 1040 ms. An integrated model was devised after combining the static and dynamic models. Experiment 2 testified the integrated model in another handwriting recognizer and found that it could still fit human performance data with users in three different training conditions. Implications of the integrated model are that: (1) when recognition accuracy and number of inputting characters are constant, the weights of average writing time for each character, segmentation time, recognition time in determining task completion time are equal but bigger than the weight of the repairing time; (2) when the repairing time, average writing time for each character, segmentation time and recognition time are constant, there is an inverse model between task completion time and recognition accuracy; when recognition accuracy is from 50% to 93%, every 1% increase of recognition accuracy will reduce task completion time from 1989 to 1915 ms; when recognition accuracy increases from 94% to 100%, every 1% increase of recognition accuracy will reduce task completion time from 1895 to 1392 ms. Guidelines in designing these recognizers were given based on these implications.