Learning Artificial Grammars with Competitive Chunking
Abstract
When exposed to a regular stimulus field, for instance generated by an artificial grammar, subjects unintentionally learn to respond efficiently to the underlying structure: Miller (1958) reports that subjects memorize letter strings generated by an artificial grammar faster than randomly generated strings. Reber (1967) reports that, following rote memorization of exemplar sentences, subjects efficiently discriminate grammatical from non-grammatical strings. We explored the hypothesis that the learning process is chunking and that grammatical knowledge is implicitly encoded in a hierarchical network of chunks. Grammatical judgments are then based on the degree to which integrated representations of strings can be built using those chunks. We trained subjects on exemplar sentences while inducing them to form specific chunks. Their grammatical knowledge was then tested with a discrimination task. We found that subjects were less sensitive to grammatical violations that preserved their chunks than to violations that did not. We derived the theory of competitive chunking (CC) and found that is successfully reproduces, via computer simulations, both Miller's experimental results and our own. Keywords: Unintentional learning, Artificial grammars, Chunking, Perception(Psychology).
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 06, 1989
- Accession Number
- ADA219270
Entities
People
- Emile Servan-schreiber
- John R. Anderson
Organizations
- Carnegie Mellon University