Seminar Team Language 08/06/2023scalab Manifestations scientifiques Séminaire
Emilie Ginestet, IGR University of Grenoble
Presentation in french with slide in english
Title: “BRAID” models: from visual word recognition to learning to read in a single-route model
For more than 40 years, dual-route models have been a reference in the study of the cognitive mechanisms involved in reading and its learning. These models mainly focus on phonological processes and postulate that phonological processing is the key to the development of reading skills. Nevertheless, these conceptual models do not describe how the mechanisms interact with each other. Computational modeling appears to be a methodological approach that is particularly well suited to answer this question.
For instance, models that implement the dual-route hypothesis postulate that sub-lexical processing (decoding) is carried out by applying grapheme-phoneme conversion rules. However, they implement a minimalist version of the visual mechanisms associated with visuo-orthographic processing.
Sylviane Valdois and Julien Diard's team has proposed a new and innovative alternative in the field by developing a series of computational models. Initiated by the development of the visual word recognition model BRAID (for Bayesian word Recognition using Attention, Interference and Dynamics), the project has recently led to the first model of reading acquisition whose single-route structure includes, among others, visual and phonological attention modules. In particular, a visuo-attentional exploration mechanism allows to simulate the transition from serial reading at the beginning of learning to a more global reading of the expert reader.
In this presentation, I will describe the general structure of the BRAID model as well as the mechanisms and sub-models implemented in later versions. Through a few examples, I will illustrate the model's ability to simulate the processing of novel words in various situations. I will emphasize the interest of these models and the development perspectives envisaged to expand the BRAID models’ family.