A team of researchers from MIT and the University of Lyon published an innovative study on dysorthography in 2024.
It presents an artificial intelligence system capable of detecting spelling errors linked to this disorder, with unprecedented accuracy.
This breakthrough could transform the identification and support of writing difficulties.
An AI trained on thousands of pupils’ written work
The researchers compiled a corpus of more than 50,000 school texts, written by dysorthographic and neurotypical children.
The AI was trained to identify recurring error patterns:
omitted letters, reversals, phonological confusions or incorrect word segmentation.
Thanks to Transformer-type language models, similar to those used in voice assistants, the system analyses each sentence in light of its grammatical context.
As a result, it can distinguish a random mistake from an error typical of dysorthography.
“Our aim is not to grade, but to understand how the brain spells”, explains linguist Dr Marion Keller, co-author of the study.
Automated yet explainable detection
One of the major strengths of this AI is its transparency.
Each detection is accompanied by a visual explanation: the algorithm indicates which letters or phonetic combinations triggered the alert.
This “explainable AI” mechanism allows teachers and speech therapists to visualise the nature of the disorder.
The results are impressive.
On a panel of 5,000 pupils, the AI recognised 92% of the errors characteristic of dysorthography, while avoiding false positives.
It also made it possible to classify profiles according to the dominant type of error: phonological, morphological or visual.
A valuable aid for speech therapists and schools
This technology does not replace diagnosis, but it offers an early screening tool.
Teachers could use it from primary school onwards to identify children at risk.
Speech therapists, for their part, would benefit from more precise monitoring of progress.
The researchers are already working on a version integrated into educational platforms.
The idea is to create a system capable of assessing the child’s actual spelling level without penalising them, and of adapting exercises automatically.
Solutions such as Dys’tap.io or AidToi.fr could soon offer these features in their training modules.
What this breakthrough changes
Dysorthography is often confused with simple spelling mistakes.
Thanks to artificial intelligence, it becomes possible to tell the disorder apart from a lack of attention.
This objective approach, based on concrete data, helps to recognise progress and avoid stigmatisation.
By analysing the reasoning rather than the result, the researchers place understanding of the disorder at the heart of teaching.
It is a new way of helping children progress while respecting their pace and their cognitive profile.
In summary
The MIT and Lyon study shows that the alliance between AI and dysorthography opens a new era for the detection of writing disorders.
With accuracy above 90%, these tools offer valuable support to families and professionals.
The future of adapted spelling support may well lie in data science.
