Is the connectionist notion of subconcepts flawed?

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The issue of mental representation of symbols has plagued cognitive science for decades and still doesn't have a resolution. The basic dispute is between the symbol system hypothesis of artificial intelligence, whose proponents include Newell and Simon [1], Newell [2], Smith [3], Fodor and Pylyshyn [4] and others, and Smolensky style connectionism [5] where a "high level" concept or symbol is represented by a number of subconcepts (subsymbols). Both sides claim their representational systems to be at the cognitive level, which means the elements used in their representational systems have meaning. In this paper, we take a closer look at Smolensky style connectionism and find it to be flawed in a number of ways. In particular, the cognitive level subconcepts (subsymbols with meaning) used by Smolensky are inconsistent with a number of principles of human learning. We argue that the subsymbolic distributed representation at the non-cognitive neural layer (McClelland and others [6], [7], [8]) is sufficient to represent a concept or symbol and that an additional layer of cognitive level subconcepts (Smolensky [5]) is redundant.

Original languageEnglish (US)
Title of host publicationProceedings of the International Joint Conference on Neural Networks
DOIs
Publication statusPublished - 2010
Event2010 6th IEEE World Congress on Computational Intelligence, WCCI 2010 - 2010 International Joint Conference on Neural Networks, IJCNN 2010 - Barcelona, Spain
Duration: Jul 18 2010Jul 23 2010

Other

Other2010 6th IEEE World Congress on Computational Intelligence, WCCI 2010 - 2010 International Joint Conference on Neural Networks, IJCNN 2010
CountrySpain
CityBarcelona
Period7/18/107/23/10

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Cite this

Roy, A. (2010). Is the connectionist notion of subconcepts flawed? In Proceedings of the International Joint Conference on Neural Networks [5596954] https://doi.org/10.1109/IJCNN.2010.5596954