Poster A18, Tuesday, August 20, 2019, 10:15 am – 12:00 pm, Restaurant Hall
Translating computational modeling into clinical practice: BiLex as a tool to simulate treatment outcomes in bilingual aphasia
Claudia Penaloza1, Uli Grasemann2, Maria Dekhtyar1, Risto Miikkulainen2, Swathi Kiran1;1Boston University, 2The University of Texas at Austin
Background. Bilinguals with aphasia (BWA) present varying degrees of impairment and recovery in their two languages. Thus, identifying the language that should be targeted in treatment is a current challenge in clinical practice. Computational models that accurately simulate rehabilitation outcomes in BWA may help predicting individual response to therapy provided in each language. Aims. We used BiLex (Peñaloza et al., under review), a computational model that can simulate L1 and L2 naming ability in healthy bilinguals, to simulate treatment effects on the treated language, and cross-language transfer effects on the untreated language in BWA. Behavioral methods. We employed a retrospective behavioral dataset of 13 Spanish-English BWA (mean age = 55.61 years) reported elsewhere (Kiran et al., 2013). All BWA received naming treatment based on semantic feature analysis in English (n = 6) or Spanish (n = 7). Individual scores on naming probes for treated words and untreated translations were collected to measure single session effects and were used as the target for simulations. Effect sizes were also computed to determine if treatment effects were significant in each language. Computational approach. First, an individual instance of the BiLex model was trained to simulate prestroke naming abilities using each participant’s age at testing, L2 age of acquisition and L1 and L2 prestroke exposure and use as model training parameters. Next, each BiLex model was lesioned to simulate L1 and L2 naming deficits as measured by standardized language tests in each BWA. Finally, each BiLex model was retrained to simulate treatment outcomes in the treated and the untreated language. Each treatment session received by a BWA was replicated as a retraining cycle for the corresponding BiLex model. After each retraining cycle, the model’s naming performance was tested in each language to measure retraining effects in both languages as done behaviorally with each BWA. Simulated naming performance in each language was then compared to the actual treatment effects for each BWA using cross-correlations. Results: Treatment gains in the treated language were significant for 10 BWA. Three BWA also showed significant transfer effects to the untreated language. Cross-correlations between behavioral treatment and computational simulation times-series data ranged between 0.48 and 0.96 for the treated language and between -0.15 and 0.63 for the untreated language. Conclusions. Overall, our results indicate that BiLex could simulate therapy effects in the treated language for most BWA, and transfer effects to the untreated language when those were observed in the BWA. These findings support the potential of BiLex to predict individual treatment outcomes in BWA by comparing overall simulated treatment effects when treatment is provided in one versus the other language. Future research with BiLex could inform clinical decisions on the language that should be targeted in treatment to observe maximum gains in BWA.
Themes: Computational Approaches, Language Therapy
Method: Computational Modeling