Conception et réalisation d'un logiciel de soutien phonétique automatisé et dégressif pour les personnes dyslexiques, allophones ou présentant des incapacités intellectuelles

Abstract

Literacy is essential for civic engagement and social participation. Yet a significant proportion of French-speaking Canadian citizens are illiterate or experience great difficulty in understanding written French. Whether it is because their native language differs from the one used in their living environment, or because of a Learning or intellectual disability, major hurdles exist in gaining access to reading and to information for these groups of people. Indeed, reading French requires skills in interpreting a significant number of unstable correspondences between graphemes and phonemes. Due to the specificities of French spelling, in which the pronunciation of a word is based on the ability to associate graphemes with phonemes, a large number of people find themselves at a disadvantage in their educational or professional setting. Thus, from a method initially developed by the teacher Dominique Rolland, the Groupe DÉFI Accessibilité (GDA) has put forward a method for textual adaptation: phonetic support. Just as training wheels are mounted on a child’s bicycle for his or her first rides, phonetic support is a mechanism that helps reduce the complexity of reading in French. It entails superimposing archigraphemes (the simplest written form of a sound) above the graphèmes that the reader has difficulty associating to the phonèmes and which potentially place them at a disadvantage. For example, the archigrapheme “o” would be written above the graphèmes “eau” and “ault”. In addition, silent letters are faded to grey. By easing the interpretation of complex graphemes of written French, this adaptation method simplifies the task and allows certain groups of people to acquire the literacy skills needed to become autonomous. Yet, the manual process of adding phonetic support to documents in Microsoft Word format requires considerable effort. Considering that the Groupe Défi Accessibilité wishes to produce educational textbooks adapted with phonetic support, such effort is too costly to ensure the viability of the project: 416 hours to adapt a textbook. Consequently, this research aims to determine if it is possible to automate this task with an error rate below 10%. Furthermore, this research must establish whether a file management system can allow for the inclusion of degressive phonetic support, which would decrease the density of phonetic support included in the texts according to the person’s skills in associating graphemes to phonemes. By using a training corpus comprised of ten texts from the grade 3 textbook Signet, adapted with phonetic support by the GDA as part of Mrs. Judith Beaulieu’s thesis, the rules of interpretation of a phonetic engine and of different components of a system have been refined and adapted. A second corpus of seven texts, also drawn from the Signet textbook (grade 3) and supplied by Mrs. Judith Beaulieu, was then utilized to validate the error rate and the categories of the software’s phonetic interpretation model. Software engineering and system architecture design tools have made it possible to build a prototype integrated into a new Microsoft Word plug-in module. The results indicate that phonetic support can be added automatically with an error rate of 2.56%, all the while optimising the operation time, since conventional manual processing time is estimated at 15 minutes per page.

Date	30 Apr 2015
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Pierre Bourque (Supervisor) & Jacques Langevin (Co-supervisor)