Oops! Silly me! Errors in a Handwriting Recognition- based Text entry Interface for Children. Janet C Read Department of Computing University of Central Lancashire Preston PR1 2HE 01772 893285 jcread@uclan.ac.uk Stuart MacFarlane Department of Computing University of Central Lancashire Preston PR1 2HE 01772 893291 sjmacfarlane@uclan.ac.uk Chris Casey Department of Computing University of Central Lancashire Preston PR1 2HE 01772 893278 ccasey@uclan.ac.uk ABSTRACT This paper describes an empirical study in which children aged 7 and 8 used handwriting recognition software and hardware to input their own unconstrained text into the computer. The children were observed using the software, and the behaviour of both the children and the system is described. Handwriting recognition is a ‘disobedient’ technology; that is, it behaves erroneously, sometimes failing to generate correct representations of the child’s intentions. This presents problems for the child, and these problems, and the strategies which the children adopted, are considered. Previous work on error correction with disobedient interfaces is used to provide grounding for the discussion. Two models are proposed, one describing user-states, the second introducing the notion of ‘tidal’ error repair. These models are then used to suggest some strategies for the design of more usable handwriting recognition interfaces for children. Keywords Text Entry, Handwriting Recognition, Usability, Errors, Children INTRODUCTION This work is part of a larger study which is looking into the efficacy of handwriting recognition as a text entry method for young children. Young children seem particularly suited to handwriting interfaces. They have difficulties with the standard QWERTY keyboard, which can make text entry laborious and can cause them to lose their train of thought when using a keyboard as a composition tool (Read et al., 2000). There appears to be some evidence that children may write more fluently when using the handwriting tablet (Read et al., 2001a). Research has indicated a shortage of work on the usability of handwriting recognition for this type of application. Handwriting, speech and gesture recognition can be collectively termed ‘disobedient interfaces’. This definition, first published in Snape et al. (1997), is used to describe recognition interfaces, which by virtue of their design have a propensity for error. These interfaces share certain characteristics; and work, which has been based on a study of one type of disobedient interface, can often inform the design of the others. This paper begins with a discussion of some of the metrics which are used to evaluate usability with handwriting interfaces, referencing relevant work, and it then goes on to describe work which has been done, both with speech and handwriting, on error handling. An empirical study is then described which looked at the patterns of errors in handwriting recognition with children. Results from this study are then presented and discussed. The conclusion offers two models which are intended to inform design, and some suggestions are made for improvements in error handling. USABILITY OF HANDWRITING INTERFACES Previous studies on the usability of handwriting recognition have focused on the rate of recognition. Frankish et al., (1995) reported recognition rates as percentage correct figures and also recorded user satisfaction using questionnaires. The mean recognition rate in their study was 87% and correlations between recognition accuracy and user satisfaction were investigated. It was found that users were more satisfied with better rates of recognition. MacKenzie and Chang, (1999) compared two handwriting recognizers by logging entry speed and accuracy. Accuracy was measured using percentage correct measures, and 87% - 93% accuracy was reported. In this test users copied words that were presented to them. Speed was expressed as wpm where 1word = 5 keystrokes or key taps. Users were given satisfaction questionnaires that similarily indicated that users were happier with high recognition rated processes. A study by the authors used children as subjects and measured accuracy, speed and user satisfaction (Read et al., 2001b). In this case, characters per second were logged for time, CER (Character Error Rate) for errors and a number Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage, and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. NordiCHI 9/02 Århus, Denmark © 2002 ACM ISBN 1-1-58113-616-1/02/0009…$5.00 Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage, and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. NordiCHI 10/02 Århus, Denmark © 2002 ACM ISBN 1-1-58113-616-1/02/0010…$5.00 35 NordiCHI, October 19-23, 2002 Papers