How computational models may help develop tailored solutions for bilingual aphasia

 

Although quite a lot is known about monolingual aphasia and its treatment options, bilingual aphasia research has been lagging behind and has only recently become the subject of systematic investigation.  As a result, formalized accounts of treatment scenarios and outcomes are few and display a huge amount of variability.  In Faroqi-Shah et al.’s recent meta-analysis for example, some of the studies reviewed were primarily naming therapies, whilst others aimed at improving sentence production and still others were more globally directed at improving communication abilities. In addition to this, as a group of patients, bilingual aphasics vary greatly along dimensions such as age of acquisition, pre-stroke proficiency in L1 and L2 and post-stoke impairment in each of the two languages. According to a recent study by Kiran and colleagues “the factors that influence treatment outcomes are not well understood. Static factors, such as pre-stroke language state, the etiology of aphasia, and level of impairment between the two languages as well as dynamic factors, such as treatment methodology, and current language exposure, add to the complicated portrait of bilingual aphasia rehabilitation.”

Kiran et al. suggest that this inherent problem of variability that has been plaguing bilingual aphasia research may be overcome by the development of computational models with the ability to simulate a wide range of individual cases.  Indeed, computational modeling allows each individual patient to be represented by a separate, individual instance of the model. The basic idea is that each one of these models will have the same architecture but will be initialized to fit the particular patient’s language history and impairment profile. In their study, Kiran et al. have run a behavioral treatment study in parallel with a computational modeling study to see if the model was able to simulate the patients’ profile, from their pre-morbid language state to their post-stroke state of impairment and finally to the outcome of the rehabilitation treatment they underwent. A more detailed and technical explanation of the architecture of the bilingual DISLEX model can be found here. For the full account of how DISLEX was trained, lesioned and retrained to simulate the individual patients in Kiran et al.’s behavioral study, I recommend reading the full article.

The great advantage of these models is that they not only simulate the bilingual aphasic’s performance after a specific course of language therapy, they can also be used to predict the outcome of alternative treatment scenarios. Kiran and colleagues are hopeful that these models could be used in the future to design individually tailored treatment plans that would result in better recovery than is currently possible.