Feedback as a coindexing mechanism in connectionist architectures

  • Authors:
  • Mark A. Jones

  • Affiliations:
  • AT&T Bell Laboratories, Murray Hill, New Jersey

  • Venue:
  • IJCAI'87 Proceedings of the 10th international joint conference on Artificial intelligence - Volume 2
  • Year:
  • 1987

Quantified Score

Hi-index 0.00

Visualization

Abstract

Co-occurrence constraints play an important role in rule-based systems such as natural language processing. The constraints appear in many forms including agreement restrictions (e.g., number agreement between subjects and predicates), selectional restrictions on complement types, and filler-gap movement dependencies. A general account of such constraints is given in a parallel execution model for rule-based systems -- Active Production Networks (APNs). The APN model is similar to connectionist (or spreading activation) models, but explicitly provides a functional interpretation of rule-based phenomena such as variable binding, multiple instantiations (including recursion), and contextual expectations. Co-occurrence constraints are represented by a theory of coindexing and trace capture, based on a feedback mechanism. Several examples of constraint processing are presented which, surprisingly, also include phenomena such as phonological nulls and contraction.