Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Rethinking eliminative connectionism

G F Marcus1

  • 1Department of Psychology, New York University, NY 10003, USA. gary.marcus@nyu.edu

Cognitive Psychology
|January 20, 1999
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Language acquisition in the absence of explicit negative evidence: can simple recurrent networks obviate the need for domain-specific learning devices?

Cognition·1999
Same author

Rule learning by seven-month-old infants.

Science (New York, N.Y.)·1999
Same author

Can connectionism save constructivism?

Cognition·1998
Same author

German inflection: the exception that proves the rule.

Cognitive psychology·1995
Same author

The acquisition of the English past tense in children and multilayered connectionist networks.

Cognition·1995
Same author

Children's overregularization of English plurals: a quantitative analysis.

Journal of child language·1995
Same journal

Sublexical semantic decoding during incidental novel word learning in natural Chinese reading.

Cognitive psychology·2026
Same journal

Seeing, hearing, and feeling causation.

Cognitive psychology·2026
Same journal

Separating decision and motor contributions to behavioral biases induced by manipulating stimulus probability.

Cognitive psychology·2026
Same journal

Congruency drives "conflict adaptation" independent of conflict: Converging evidence from behavior and computational modeling.

Cognitive psychology·2026
Same journal

Corrigendum to "Network analyses identify critical factors for facilitating future-oriented decision-making" [Cogn. Psychol. 165 (2026) 101815].

Cognitive psychology·2026
Same journal

The time course of local coherence effects in German: Evidence from self-paced reading times and event-related potentials.

Cognitive psychology·2026
See all related articles

Eliminative connectionist models struggle to generalize universal rules to new examples. Current models fail to extend learning beyond training data, unlike human reasoning which handles arbitrary instances.

Area of Science:

  • Cognitive Science
  • Artificial Intelligence
  • Linguistics

Background:

  • Humans excel at generalizing universal rules to novel situations.
  • Two main theories explain this: symbol manipulation and connectionist (eliminative) models.
  • The efficacy of eliminative connectionist models in generalizing universals is debated.

Purpose of the Study:

  • To investigate whether eliminative connectionist models can account for the generalization of universals to arbitrary instances.
  • To determine if current eliminative connectionist models can extend learning beyond their training data.

Main Methods:

  • Analysis of the theoretical capabilities of eliminative connectionist models.
  • Examination of the generalization limitations of popular eliminative connectionist architectures.

Related Experiment Videos

  • Comparison of model performance with human generalization abilities.

    • Main Results:

    • Eliminative connectionist models, as currently conceived, cannot generalize universals to arbitrary items outside their training space.
    • Popular eliminative connectionist models fail to learn generalizations beyond the scope of their training examples.
    • Human generalization of universals appears to require mechanisms beyond current eliminative connectionist capabilities.

    Conclusions:

    • Eliminative connectionist models face significant challenges in replicating human-like generalization of universals.
    • Current popular eliminative connectionist models are insufficient for explaining how humans apply rules to novel instances.
    • Architectures incorporating symbol manipulation may be necessary to bridge this gap in artificial intelligence and cognitive science.