Reading focus
Syllabus: How Do We Organize Knowledge?
Turn Rosch's distinction between a phenomenon and its explanation into a model comparison. Follow the verbal meaning of the equations and the transfer predictions; a typical-looking response can arise without consulting an abstract prototype.
A question to consider. Which training or transfer comparison would distinguish the context model from the independent-cue accounts tested, even if both predict a general advantage for typical items?
Why this reading
A new object can resemble a category's average yet be unlike its particular members. Medin and Schaffer make that difference testable: their model classifies by similarity to remembered instances. The paper shows why explaining typicality requires more than observing that some examples are easier to classify.
Before you read
An exemplar is an individual encountered example; a prototype summarizes a category. In this paper, a cue is a stimulus attribute, such as shape or color. Independent-cue models combine separate attribute contributions. The context model makes combinations of attributes matter through their joint similarity to stored examples. Transfer trials ask participants to classify new items after training.
Read for
- Translate the model's operations into words: compare a probe with stored examples, combine similarity across attributes, and accumulate support for each category.
- Use the training and transfer patterns to identify predictions that distinguish the context model from the particular independent-cue models tested.
- Separate the claim that classification can use specific instances from the stronger claim that people never acquire abstract knowledge. Notice what the model leaves unspecified about learning and storage.
Consider afterwards
- Construct two training sets with the same average features but different combinations of features. What new item could help distinguish a summary-based judgment from one based on the remembered combinations?
Editorial context
This is a formal model and experimental comparison, not a demonstration that every category is stored only as exemplars. Its success depends on the comparison models, similarity assumptions, training materials, and task. Smith and Minda's later study of learning trajectories offers a useful counterpoint: support for prototype and exemplar descriptions can change across training and category structures. A pattern called a prototype effect does not uniquely identify a prototype representation.