Abstract

Human cognition is said to be systematic: cognitive ability generalizes to structurally related behaviours. The connectionist approach to cognitive theorizing has been strongly criticized for its failure to explain systematicity. Demonstrations of generalization notwithstanding, I show that two widely used networks (feedforward and recurrent) do not support systematicity under the condition of local input/output representations. For a connectionist explanation of systematicity, these results leave two choices, either: (1) develop models capable of systematicity under local input/output representations; or (2) justify the choice of similarity-based (nonlocal) component representations sufficient for systematicity.

Citations

2304	Learning Internal Representations by Error Propagation - Rumelhart, Hinton, et al. - 1986
1530	Introduction to the theory of neural computation - Hertz, Krogh - 1991
1313	Finding structure in time - Elman - 1990
488	Connectionism and cognitive architecture: a critical analysis - Fodor, Pylyshyn - 1988
333	Rethinking innateness. A connectionist perspective on development - Elman, Bates, et al. - 1996
299	Recursive distributed representations - Pollack - 1990
296	On the proper treatment of connectionism - Smolensky - 1988
214	Tensor product variable binding and the representation of symbolic structures in connectionist systems - Smolensky - 1990
156	Serial Order: A parallel Distributed Processing Approach - Jordan - 1986
140	Mechanisms of sentence processing: assigning roles to constituents - McClelland, Kawamoto - 1986
107	Syntactic transformations on distributed representations. Connection Science 2(1–2):53–62 - Chalmers - 1990
102	Mapping part-whole hierarchies into connectionist networks - Hinton - 1990
99	Learning and applying contextual constraints in sentence comprehension - John, McClelland - 1990
82	Learning from Hints in Neural Networks - Abu–Mostafa - 1990
59	Incremental learning, or the importance of starting small - Elman - 1991
54	Connectionism and the problem of systematicity: Why Smolensky’s solution doesn’t work - Fodor, McLaughlin - 1990
52	Representation and structure in connectionist models (Technical Report CRL-8903 - Elman - 1989
39	Systematicity in connectionist language learning - Hadley - 1994
32	Training second-order recurrent neural networks using hints - Omlin - 1992
26	Generalization and connectionist language learning - Christiansen - 1994
19	Connectionism, Constituency, and the Language of Thought - Smolensky - 1991
18	Constituent structure and explanation in an integrated connectionist/symbolic cognitive architecture - Smolensky - 1995
17	Compositionality and systematicity in connectionist language learning - Hadley - 1993
16	Can connectionist models exhibit non-classical structure sensitivity - Niklasson, Gelder - 1994
16	Exercises in rethinking innateness. A handbook for connectionist simulations - Plunkett, Elman - 1997
13	Strong systematicity within a connectionist framework: The tensor-recurrent network - Phillips - 1994
10	Structure sensitivity in connectionist models - Niklasson - 1993
8	Systematicity: Psychological evidence with connectionist implications - Halford - 1997
6	Connectionism and the problem of systematicity (continued): Why Smolensky's solution still doesn't work - Fodor - 1997
6	Strong semantic systematicity from unsupervised connectionist learning - Hadley - 1995
5	Connectionism and the problem of systematicity - Phillips - 1995
4	Systematicity and connectionist language learning - Niklasson, Gelder - 1994
4	Classicism and cognitive architecture - Gelder - 1994
2	Forcing simple recurrent networks to encode context - Maskara - 1992
1	New Jersey Institute of Technology, NJ. Also - Rep
1	Perceptrons (4th edition - Minsky - 1990