CHS: Small: Early Dyslexia Detection and Support at Scale to Help Students Succeed in School

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

At least 10% of the population has dyslexia, which results in difficulty with reading and writing, and often leads to school failure (40% of those who drop out of school have dyslexia). If people know they have dyslexia, they can with effort train over time to overcome its negative effects. Yet even though we know how to detect dyslexia, most children are diagnosed late because current procedures are expensive and require professional oversight. The PI's goal is for everyone to know as early as possible if they might have dyslexia; his approach to achieving this goal is to make it easy, inexpensive, and even enjoyable to find out. To these ends, he and his team plan to design personalized game activities based on the detection results to target the cognitive skills with which students need to practice most. Much of the research into detection and support activities has thus far taken place in the lab; the team plans to extend this work into the real world via wearable devices to help people with dyslexia better read and write in the context of learning activities outside of the classroom, e.g., in museums or at historic sites. The work will build on the team's Dytective software, which the PI plans to publicly release along with other tools that will be developed and refined as part of this research. The team will work with community partners like dyslexia organizations and schools, undergraduate and graduate students, and experts from related fields to disseminate their findings as widely as possible, to nurture the development of young researchers in this area, and to integrate their work with other related efforts. In addition, new course modules on Dyslexia and Language Technology will be added to the Human Factors course at CMU. Current approaches for detecting dyslexia require either a professional psychologist or expensive brain imaging equipment (and an expert to run it and interpret the results). The PI's approach to detecting and supporting dyslexia uses a scalable web-based game. The method relies on human-computer interaction metrics drawn from people playing games designed with a linguistic and empirical understanding of the errors that people with dyslexia tend to make. Machine learning over this data may allow for the detection of dyslexia much earlier (and at much less expense) than would otherwise be possible. Although the current research is informed by prior work characterizing the origin of dyslexia and its linguistic manifestation, the intellectual contribution lies instead in understanding how we can use data from game play to detect dyslexia. This approach, once demonstrated, may generalize to other areas, and the exercises that are found to be most useful in detecting or supporting dyslexia may also inform our basic understanding of dyslexia.