Bilingual aphasia rehabilitation faces the challenge of determining which language to target in therapy to maximize recovery across both languages. This double-blind randomized controlled trial (48 Spanish–English bilinguals with chronic aphasia; NCT02916524) evaluated whether the BiLex computational model could predict the optimal language for aphasia therapy. Participants received 40 h of semantic feature-based treatment in either the BiLex-recommended language or the opposite language. Both groups showed similar gains in treated-language naming, with no significant difference in proportion of maximal improvement (Difference (SE) = –0.03 (0.07); t = –0.46; p = 0.65). However, the model-opposite group showed significantly greater cross-language generalization (Difference (SE) = –0.16 (0.07); t = –2.38; p = 0.02), though with higher response variability. Further, when the participants were divided into subgroups according to performance, the model-assigned group had a significant advantage in all but the lowest performing subgroups. All these differences were consistent with BiLex model predictions.

PDF

npj Digital Medicine (2026).

@article{kiran:npj26, title={Predicting Bilingual Aphasia Treatment Outcomes Using Digital Twins: A Double-Blind Randomized Controlled Trial}, author={Swathi Kiran and Erin Carpenter and Uli Grasemann and Michael Scimeca and Manuel J. Marte and Marissa Russell-Meill and Claudia Penaloza and Yorghos Tripodis and Risto Miikkulainen}, journal={npj Digital Medicine}, month={ }, url="http://www.cs.utexas.edu/users/ai-lab?kiran:npj26", year={2026} }