Handwriting Analytics

Abstract

This chapter reports the development of a project for analyzing, modeling and remediating handwriting difficulties. Initially built upon robotic activities, the project ended up as a method to analyze handwriting traces produced on a digital tablet. These tools capture the dynamics of handwriting, such as the changes in pen pressure, acceleration or tilt, which the eye of a therapist can’t perceive. Then, machine-learning algorithms elaborate models that predict dysgraphia. We initially used supervised learning algorithms, trained with the results of standardized paper-based writing tests manually graded by therapists. They produced a binary diagnosis: dysgraphia or not. We then moved toward unsupervised learning and replaced the binary output with a value on a gradient of handwriting difficulties. Several subtypes of dysgraphia have been identified. We also consider the transferability of our model across different languages and scripts, as well as its relationship to dyslexia and the influence of body posture.