What is involved in solving the Marcus et al. task?

• The task involves the detection of a pattern.
  • The pattern has sequential positions containing elements.
  • There are relations of similarity among the pattern elements.
  • These relations are extracted and stored somewhere.
  • (The specific content of the pattern elements is apparently ignored.)

• A symbolic way to model this behavior:
  • The child extracts a rule with variables: ?A ?A ?B, ?A ?B ?A, or ?A ?B ?B.
  • When novel sequences are presented to this model, a pattern matching mechanism is applied.
    • The mechanism treats variables completely differently from constants, and the specific content of the individual sequence elements is irrelevant.
    • Matching is all-or-none.

• Another way to model this behavior:
  • The child segments each input into groups based on similarity and learns the extracted similarity/dissimilarity relations among the elements at different positions.
  • When novel sequences are presented to this sort of model, the model attempts to segment them and then activate the learned relations.
    • The content of the individual sequence elements may be relevant because the relations may only have been learned for some elements.
    • Similarity is a matter of degree.