How many Neurons for your ‘Grandmother’ ? Three Arguments for Localised Representations

In a recent article, Bowers (2009) argues that local representations are more consistent with neuro-biological data than distributed representations, as typically generated in Parallel Distributed Processing (PDP) models. We present three reasons why localised neural representations are good candidates for supporting mental representations, as they provide a solution to the trade-off between combinatorial arguments that favour fully-distributed representations and metabolic arguments which favour localist representations.