On the potential of non-classical constituency

Horgan and Tienson doubt that classical cognitive science will be able to solve the frame problem but have some expectations with respect to connectionism. Since the frame problem arises in contexts where a potentially large amount of complex knowledge is involved, connectionism has to prove that its models can represent and use well-structured information. Indeed, Horgan and Tienson note that the key question for the viability of connectionism is whether representations with some kind of non-classical encoding of constituency are susceptible to richly structure-sensitive processing. I will examine several connectionist models and point out three reasons for doubting that work of this kind can scale up to the extent required for dealing with the frame problem. I will conclude that connectionism as of yet has no principled and satisfactory way of effectively representing structured information in a distributed way. Hence, the frame problem provides a difficulty to connectionism that is no less serious than the obstacle it constitutes for classical cognitive science.