Reading focus
Syllabus: How Do We Organize Knowledge?
Make the proposed network answer to response-time evidence. Relate its hierarchical organization to Rosch's category levels while keeping a descriptive taxonomy separate from a retrieval mechanism.
A question to consider. What assumptions must hold before a longer verification time counts as evidence for traversing more links, and how could expertise complicate that inference?
Why this reading
How could the time needed to verify a familiar fact reveal the organization of knowledge? Collins and Quillian turn a proposed hierarchy into response-time predictions. The paper makes the link between a representation and the process operating on it unusually explicit.
Before you read
Semantic memory concerns knowledge rather than recollection of a particular event. A node represents a concept in the proposed network; a link connects it to another concept or property. Cognitive economy means storing shared information at a broader level instead of repeating it for every member. Response time measures the whole task, so interpreting it requires assumptions about retrieval and decision processes.
Read for
- Connect the hierarchy in Figure 1 to the predicted time for verifying category membership and properties at different levels.
- Inspect the illustrative sentences, selected hierarchies, and response-time pattern. Separate the observed differences from the assumed time for traversing a link.
- Compare the treatment of true and false judgments, including the authors' uncertainty about how false judgments are made and their qualifications about familiarity and duplicate storage.
Consider afterwards
- If a broad category statement is verified faster than a narrower one, what would that challenge in this model, and what alternative explanation involving familiarity would you need to examine?
Editorial context
This compact historical experiment uses small samples and deliberately selected, relatively orderly hierarchies. Eight people participated in each experiment, but equipment failure left usable data from only three in the third. Its response times constrain a model under stated assumptions; they do not provide a literal map of the brain or establish that all knowledge forms a strict hierarchy. Collins and Loftus's later spreading-activation account extends the network approach to semantic relatedness and typicality.