Modeling orthographic learning of new words in BRAID, a Bayesian model of visual word recognition.

Ginestet, E. . , Valdois , S. & Diard, J. .

Univ. Grenoble Alpes, CNRS, LPNC UMR 5105, F-38000 Grenoble

We present an extension of BRAID, a Bayesian model of expert word recognition, to model the acquisition of new orthographic knowledge. To a classical three-layer architecture, BRAID adds an original attentional component, allowing to control how attention is deployed on the word stimulus.

To model orthographic learning, we assume that visual attention is distributed over the letter string so as to optimize the accumulation of perceptual information about letters, to construct efficiently a new orthographic memory trace. Furthermore, we assume that letter perception is influenced by lexical knowledge, in a top-down manner modulated by the probability that the stimulus presented is a known word.

In this study we simulate observations from an eye-tracking experiment involving repeated reading and implicit learning of thirty French new 8-letter words. Results show that the model successfully reproduces the decrease of the number of fixations and of processing time we observed along repetitions.