Subject: Automatic generation of constrained text for the detection and monitoring of visual pathologies: Development of fundamental AI methods in natural language processing
Supervisors: Pierre Kornprobst (Biovision project-team), Jean-Charles Régin (Université Nice-Sophia Antipolis and I3S, CNRS), and Aurelie Calabrese (Biovision project-team)
Abstract: Because reading involves many functions, perceptual, oculomotor, and cognitive, assessing the way we read is a formidable indicator of our abilities and possible dysfunctions of these functions. This thesis’s objective is to focus on visual impairments. The evaluation of reading performance tends to become one of the essential clinical measures to diagnose the pathology, to judge the effectiveness of treatments, surgical procedures, or rehabilitation techniques. The difficulty is to have enough equivalent texts in linguistic complexity to avoid any bias in the measurements. A typical example that we have chosen as a starting point is the MNREAD test – a standardized reading test used worldwide in clinical and research contexts. To construct their sentence sets, the authors used well-defined, strong constraints. This test is available in 19 languages, but MNREAD sentences are tough to produce due to their very restrictive nature. Only a minimal number of test versions are available (only two in French, i.e., 38 sentences), limiting the use of this test in clinical settings. Therefore, this thesis’s objective is to design methods for the automatic generation of text under constraints to overcome these limitations.
For more information and to apply: click here
