Computational Models of Aphasia

Computational models provide a bridge between theories and data. We can test our theoretical assumptions by implementing them in a computer program, and observing whether it can reproduce the observed data. We can also use the model to generate predictions about data we have yet to observe.

To explore the implications of our theory of the auditory system’s involvement in speech production, we have created a computational model of picture naming, which can be applied to data from healthy and aphasic speakers. The model can take counts of response types, including correct and five other error types, and return estimates of localized connection strengths within the lexical network. Alternatively, the connection strengths can be specified, and the model can produce a distribution of naming responses.

Our theoretical assumptions are embodied in the structure of the lexical network. By including auditory representations in a model of the naming process, we believe we can identify specific behavioral consequences that emerge in the context of aphasia.

Our new model can be compared with another popular model on our website.

http://www.cogsci.uci.edu/~alns/webfit