Abstract:

We report on a follow-up study of the Co-Writer project at EPFL [1]; we confirm their findings, extend the applicability to another language with a different alphabet (Greek) and go into an in-depth qualitative study of the child-robot relationship. The core idea of “learning by teaching” is that the student learns by undertaking to teach what they are supposed to learn which motivates them into ownership of the process. Although a highly effective method [2], there are serious practical obstacles in scaling learning by teaching to larger numbers or extending it to younger learners and to certain skills such as handwriting. This is where we enlist the assistance of a humanoid robot to help a young learner with her handwriting by assuming the role of the bad-writer pupil who asks the help of his human friend. The robot pretends writing mistakes similar to the child’s own; the child recognizes the errors as such even when she would refuse to acknowledge them in her own writing, and makes an extra effort for better writing as the robot’s instructor; the robot improves its writing; the child goes through another round of amelioration; and, hopefully, gets in the habit of better writing.

In order to extend the original study for the Greek language and alphabet we first needed to collect data from beginner writers, children of 6-9 years old. The nine-month study for collecting and evaluating handwriting samples from primary school students in Cyprus in order to examine and describe dysgraphia for the Greek alphabet used a questionnaire, an evaluation sheet for the letter formation and copies of students’ notebooks. It involved four primary schools in Limassol, Cyprus and a total of 68 students and 11 teachers. We were thus able to obtain a consensus on what constitutes poor letter writing in Greek. All writing samples were evaluated by teachers and were classified based on Chandra et al. [3] taxonomy which was appropriately extended.

The main experiment involved three case studies of children who held three sessions each with the humanoid programmable robot NAO. We video recorded each child’s reactions and their progress in handwriting. In all three cases, albeit not in the same degree, we saw that (a) the child-robot interaction quickly improved; the increasing number of correct remarks from all children on only two meetings shows the comfort that they gained with lab environment and the robot itself and (b) the correctness of children remarks and their comments about letters shows their ability to distinguish the correct from the wrong letters and to justify their answers. Both conclusions confirm the finding of the Co-Writer project [4]

References:
[1] Jacq etal, Building Successful Long Child-Robot Interactions in a Learning Context, 11th ACM/IEEE International Conference on Human-Robot Interaction (HRI), Christchurch, NEW ZEALAND, MAR 07-10, 2016
[2] Duran D., Learning-by-teaching. Evidence and implicationsas a pedagogical mechanism, Innovations in Education and Teaching International, Routledge, 2016
[3] Chandra etal (2017) Classification of Children's Handwriting Errors for the Design of an Educational Co-writer Robotic Peer. In Proceedings of the 2017 Conference on Interaction Design and Children, ACM, 215-225.
[4] Lemaignan etal, From real-time attention assessment to with-me-ness in human-robot interaction. In 11th ACM/IEEE International Conference on Human Robot Interaction (pp. 157-164)

Keywords:

Dysgraphia in Greek, primary education, educational technology, human-robot interaction.