Computational Modeling of Bilingual Aphasia Rehabilitation

Here is a recent presentation that summarizes our current work on bilingual aphasia rehabilitation and fMRI studies. Bilingual Aphasia

Simulating Bilingual Aphasia Rehabilitation: A computational Model

Current research on bilingual aphasia has only begun to inform us about the optimal rehabilitation for bilingual aphasic patients (Roberts & Kiran, 2007; Edmonds & Kiran, 2006) but the literature is still sparse in terms of interpreting the nature of naming impairments in bilingual aphasia. We have developed a computational model to simulate an English-Spanish bilingual language system in which language representations can vary by age of acquisition and relative proficiency in the two languages. This model is subsequently lesioned at specific sites by varying connection strengths between the semantic and phonological networks.


Based on recent theoretical models, the three self-organizing maps (semantic, L1 and L2) each with 30×40 neurons were trained simultaneously with the associative connections between each pair of maps. All three maps used Gaussian neighborhood functions whose width decreased exponentially (from sigma = 7.0 down to 0.2) over the course of training. Later, age of acquisition for L2 was simulated by delaying the onset of training for the L2 phonetic map and its associative connections. Differences in exposure to L1 and L2 were simulated by exposing the model to more or less phonetic input in each language.

In order to match the model’s performance in both English and Spanish to that of a group (N = 39) of individual bilingual human speakers with varying AoA and relative proficiency, the training parameters were set up to match the known ages of acquisition and exposure data as closely as possible for each test case.

¬†We then extended the model to simulate a group of bilingual aphasia patients (N = 19), by attempting to replicate the patients’ self-reported AoA and pre-stroke performance. In most cases the model usually comes close to the target performance in both languages indicating the validity of this model in simulating naming impairment in bilingual aphasia.

Kiran.Computational modeling


Slide70 Slide71 Slide72