Model-oriented naming therapy: Testing predictions of a connectionist model.

Background: The two versions of the connectionist model of Dell and colleagues offer alternative explanations of aphasic naming disorders (Dell, Schwartz, Martin, Saffran, & Gagnon, 1997; Foygel & Dell, 2000). The semantic-phonological (SP) model hypothesises impairments in lexical-semantic or lexical-phonological connections, and the weight-decay (WD) model assumes global impairments in either connection weights or activation decay. In each version, a patient's error pattern in picture naming is simulated to assess the underlying disorder (connectionist "diagnosis"). A systematic comparison of both model versions in model-oriented naming therapy has not yet been performed. Moreover, if the normalisation of the error pattern during recovery is lesion-specific, as suggested in the SP model (Schwartz, Dell, Martin, Gahl, & Sobel, 2006), this should be observable in the patient data. Aims: Predictions were made and tested regarding the relation between (1) connectionist diagnosis and therapy outcome, and (2) connectionist diagnosis and error pattern development. For example, patients with phonological disorders in the SP model should (1) benefit more from phonological as compared to semantic therapy, and (2) present a decrease of nonwords in their naming responses. Methods & Procedures: The connectionist diagnosis and a 4-week therapy with cueing hierarchies (Howard, 2000; Wambaugh et al., 2001) were administered to 10 German-speaking aphasic patients with naming disorders. Six patients, who had been diagnosed by the SP model, received semantic and phonological therapy. The other four patients, diagnosed by the WD model, received increasing and vanishing therapy (Abel, Schultz, Radermacher, Willmes, & Huber, 2005). Outcomes & Results: Cueing therapy was generally effective for 9 of 10 patients. The trend of improvement was always found in the direction predicted by the connectionist diagnosis, except for two patients diagnosed by the SP model who presented a numerical trend in the opposite direction. Nevertheless, the SP model offered a more plausible explanation of lesion-specific therapy outcomes, and it properly predicted the error pattern development. Moreover, the errorless learning procedure applied in vanishing therapy was favourable for patients with phonological (SP model) or weight (WD model) lesions, and this may be attributed to their characteristic error types and an impairment of editorial processes. Conclusions: Models can be informative about the effectiveness of potential therapies and error pattern developments. Data from therapy studies can test competing models. [ABSTRACT FROM AUTHOR]