Connectionist Models

Question 1

Thomas + Karmiloff-Smith 2002

Answer 1

Older models could simulate disorders (but manipulations gave similar effects). Test residual normality. Dual-route model with lesion each at start. Anything less than total lesion is adopted by the other route at the expense of performance. Training allows for compensation

Question 2

Thomas + Karmiloff-Smith 2003

Answer 2

Modelled WS for past-tense formation. Used models of atypical phonology and reduced lexical semantics. Too post-hoc?

Question 3

Harm + Seidenberg 1999

Answer 3

Replicate phonological + surface dyslexia by impairing the representation of phonological information and reduced learning rate

Question 4

Criticism

Answer 4

Have to teach it to be systemised (Fodor), models aren’t domain-general, qualitative representational change can’t occur, no shared conceptual activation

Question 5

Other models can simulate disorders, but not selective impairment. Tests residual normality using dual route. As long as not full lesion, other route adopts info at the expense of reduced performance

Answer 5

Thomas + Karmiloff-Smith 2002

Question 6

WS model using past-tense. Atypical phonology and reduced lexical semantics. Too post-hoc?

Answer 6

Thomas + Karmiloff-Smith 2003

Question 7

Replicate phonological + surface dyslexia by impairing the representation of phonological information and reduced learning rate

Answer 7

Harm + Seidenberg 1999