The story behind Dytective: how we brought research results on dyslexia and accessibility to Spanish public schools
Luz Rello · 2022 · Proceedings of the 19th International Web for All Conference (W4A) · doi:10.1145/3493612.3520442
Summary
This invited talk paper presents the decade-long research and entrepreneurial journey behind Dytective, a platform that combines machine learning and gamified exercises to detect risk of dyslexia and provide personalized interventions. Dyslexia affects up to 10% of the global population but remains under-diagnosed, particularly in languages with transparent orthographies like Spanish where it is called a "hidden disability." The author identifies three key barriers: dyslexia is unknown and under-diagnosed, it leads to school failure through reading and writing difficulties, and current diagnosis and treatment are expensive, creating socio-economic barriers. Dytective addresses these through a free online screener, linguistic support exercises, and scholarships for under-privileged families, delivered through the social venture Change Dyslexia. The screener uses a Random Forests machine learning model trained on human-computer interaction data from a gamified test with over 5,000 participants, achieving approximately 80% sensitivity or recall depending on age group. The 15-minute test is available for children aged 7-17, primarily in Spanish with preliminary English results. For treatment, the team designed 5,000 exercises based on linguistic error patterns of people with dyslexia, later expanded to 42,000 exercises personalized across 24 cognitive ability indicators that strengthen weak skills and challenge strong ones. Evaluation showed significant improvement in reading tasks after playing 20 minutes, four times per week, over four weeks.
Key findings
Dytective has been used nearly 350,000 times across 50 countries and has been adopted by over 800 Spanish public schools in collaboration with regional governments — the first large-scale dyslexia screening and support effort in the Hispanic world. The paper distills several practical lessons from this journey. Descriptive statistics from small user studies (40-60 participants) proved essential for understanding linguistic patterns before investing in large-scale ML training data. Cross-functional teams spanning linguistics, data mining, machine learning, psychology, pediatrics, practitioners, and people with dyslexia were invaluable both for research quality and for anticipating launch issues. The team switched from deep learning to Random Forests to achieve better interpretability after user criticism about unexplainable results. Sustainability required transitioning from a free research tool (Dyseggxia, which reached 32,000 users in 72 countries but was not economically viable) to a social enterprise model. Community involvement through a Kickstarter campaign with 500 backers helped shape the product before release. Corporate partnerships (Samsung) and government collaboration amplified social impact far beyond what the research team could achieve alone.
Relevance
This paper is a compelling case study of how accessibility research can scale from lab findings to real-world impact affecting hundreds of thousands of people. For accessibility practitioners and researchers, it offers practical lessons about the path from research to deployment: the importance of explainable AI when tools affect people with disabilities, the need for sustainable business models alongside social missions, and the value of involving the target community throughout development. The work demonstrates that accessibility tools can achieve massive scale when they address genuine barriers — in this case, the cost and availability of dyslexia screening. The transition from free-but-unsustainable to social-enterprise is instructive for anyone building accessibility tools. Future goals include adapting the approach to non-phonetic alphabets like Chinese and developing pre-reader assessment using non-linguistic items, potentially creating a language-independent universal dyslexia screening approach.
Tags: dyslexia · machine learning · screening · serious games · social entrepreneurship · reading accessibility · education · Spanish language