Histoire  |  Géographie  |  Économie  |  Société

 

Alice Lane, 2006 | Gryon, VD

 

Supporting dyslexics requires expensive diagnosis, challenging in multilingual settings where language skills may mask dyslexia. Elbro et al. (2012) proposed the ability to fuse symbols into words (phoneme fusion), a known challenge for people with dyslexia, as a language-independent dyslexia test. This project aimed to make the Elbro test more widely available as a web-based screening tool for use prior to professional assessment. Transferring it to a web-based setting required sound recognition, based upon Artificial Intelligence (AI), to test the ability to fuse words. Dyslexics also struggle with phoneme segmentation, not part of the Elbro test. This project added segmentation, which is not dependent on sound recognition. Quantitative testing showed that, due to unreliable sound recognition, the Elbro fusion-based test did not translate to a web setting; but that there was a significant difference (p <0.05) between test scores of diagnosed dyslexics and others for the segmentation test. This could be the basis of a web-based dyslexia screening tool and is being further validated as such.

Introduction

I was 15 when I was diagnosed with dyslexia. My struggles with French were attributed to being bilingual until a psychologist suggested testing for dyslexia, which was confirmed. I was not a failure in French. Rather, I had a learning disability.
Educational practitioners’ knowledge and ability to recognise dyslexic symptoms vary, meaning schools cannot be relied upon to encourage a costly, formal diagnosis. In a multilingual setting, common in Switzerland, problems with reading and writing may be misattributed to language rather than dyslexia. Elbro et al. (2012) developed a language-independent dyslexia test for expert use. This project initially aimed to develop the Elbro test for the internet as an accessible screening tool prior to professional diagnosis.

Methods

The Elbro test evaluates phoneme fusion. Phonemes are the sounds (e.g. /k/-/a/-/t/) we fuse to make a word (cat). Dyslexics struggle with this when reading. I added a new test using phoneme segmentation, something I find challenging as a dyslexic. Segmentation involves separating phonemes in a spoken word (/k/-/a/-/t/) when writing (cat). To make the test language independent, Elbro uses symbols and non-words instead of letters and words.
The fusion and segmentation tests were coded into a website with visual instructions and minimal text (translated into 7 languages) to make it dyslexia- and language-independent. Fusion was tested by the ability of a user to speak a non-word associated with a set of symbols, evaluated using AI-based sound recognition. Segmentation was tested by the user choosing symbols when they heard a non-word.
The web-based tool was used by 68 participants (15 dyslexic) to evaluate if it screened for dyslexia. Additional data on participants were obtained for confounding factors such as age (13–78) and languages spoken (English 62%, French 29% and 19% bilingual). A difference of medians test (McGill et al., 1978) tested for fusion and segmentation score differences.

Results

The segmentation phase distinguishes between dyslexics (D) and non-dyslexics (ND), as the median score (out of 25) was significantly (p <0.05) lower (D score 18, ND score 24). The fusion phases had no significant difference (D 12, ND 13). 50% of participants reported issues with the sound recognition required for the fusion test. NDs had fusion phase scores with very low kurtosis (K=1.7) as many scored poorly, whereas in the segmentation phase nearly all NDs performed well, giving high kurtosis (K=8.5). Poor ND performance in the fusion test is likely due to unreliable sound recognition of non-words.

Discussion

The original Elbro fusion-based test did not transfer reliably to a web-based setting with current technology. The most likely explanation is inconsistent sound recognition and whilst technological developments may improve performance, sound recognition across different microphones and accents will always be a challenge. The novel segmentation test does not require sound recognition and produced clear differences between diagnosed dyslexics and other participants. It is ideally suited for web-based use.

Conclusions

A segmentation-based phoneme test on a web platform as a dyslexia screening tool has been created and evaluated. This is now being operationalised for use, including an interactive map allowing the user to access the dyslexia association for their country. Further data collection is needed to increase the sample size, particularly younger individuals. Independent expert validation has now started.

 

 

Appréciation de l’experte

Dr. Camille Farcy

Le travail de la candidate est remarquable par sa rigueur et son approche innovante. Elle analyse les défis du dépistage de la dyslexie en contexte multilingue, combinant solutions technologiques et méthodologie expérimentale. Son étude met en lumière les limites de la reconnaissance sonore et propose une alternative efficace avec le test de segmentation. L’analyse des résultats est claire et bien argumentée, démontrant une excellente maîtrise du sujet. Ce travail contribue à améliorer les outils de dépistage et pourrait inspirer de futurs travaux dans le domaine de la recherche linguistique.

Mention:

excellent

Prix spécial «Exporecerca Jove – Barcelona Science Fair» décerné par les soutiens de Science et jeunesse

 

 

 

Lycée-Collège de St-Maurice, St-Maurice
Enseignant: Jan Schönbächler