Categorization
Plato famously characterised humans as seeking to ‘carve nature at its joints’ (Phaedrus). The rough idea is that we find ways to classify and group the things in our experience in ways that somehow respect their natural clustering and separation.
This topic looks at cognitive models that have tried to capture how and why we partition the world in the ways we do.
Primary Readings
Everyone should read these and be prepared to discuss.
| Posner and Keele (1968) | On the genesis of abstract ideas. Journal of Experimental Psychology, 77 (3).This is the original appearance of the now classic “prototype theory” of categorization. |
| Love, B. C., Medin, D. L., & Gureckis, T. M. (2004). | SUSTAIN: a network model of category learning. Psychological Review, 111(2), 309.Here’s a much more recent “non-parametric” statistical category model. Can it resolve some of the shortcomings or tensions between prototype and exemplar models. |
Secondary Readings
The presenter should read and incorporate at least two of these.
| Murphy, G. L. (1968) G. L. (2002) | Chapter 3, The big book of concepts. MIT Press (On Learn)This chapter discusses theories of concepts, specifically focusing on the prototype view, exemplar view, and the knowledge approach. The prototype view suggests that categories are represented by a prototype, which is a summary representation of the entire category with certain features being more important. The exemplar view states that categories are represented by individual exemplars that we remember and compare to new items. The knowledge approach argues that concepts are shaped by prior knowledge and understanding. The passage also introduces the Generalized Context Model (GCM), which calculates similarity and categorization decisions based on distances between exemplars. It discusses the use of weighted dimensions and the Luce choice axiom in the GCM. The GCM has received both praise and criticism but has been influential in the field of categorization. Overall, the chapter highlights the complexity of concept representation and the various factors that influence categorization decisions. |
| Medin, D., & Schaffer, M. (1978). | Context theory of classification. Psychological Review, 85, 207-238.This article discusses the theory of classification learning, specifically the context model, which proposes that judgments are based on stored exemplar information and similarity between stimuli. Four experiments consistently support the context model over other theories. The article also discusses the structure of concepts and the role of exemplars in learning. This is the original appearance of the now classic “exemplar theory” of categorization. |
| Rosch, E., Mervis, C.B., Gray, W., Johnson, D., & Boyes-Braem, P. (1976) | Basic Objects in Natural Categories. Cognitive Psychology, 8(3), 382-439.The study conducted by Rosch et al. explores the concept of categorization in cognitive psychology, specifically focusing on basic level categories. The researchers argue that categories in the concrete world are not arbitrary, but determined by the correlational structure of perceived objects and the need for cognitive efficiency. They suggest that basic categories are the most inclusive and differentiated, carrying the most information and possessing high category cue validity. Another classic. |
| Gershman, S. J. (2017) | On the Blessing of Abstraction. Quarterly Journal of Experimental Psychology 70 (3).The blessing of abstraction refers to the phenomenon that abstract knowledge can sometimes be acquired more quickly than specific knowledge. Gershman tests this hypothesis using hierarchical Bayesian models and an experiment manipulating the dispersion of variables. The experiment involved participants predicting the next button press in sequences of button presses. The size range of the buttons was manipulated to create different levels of dispersion. The results showed that in the low dispersion condition, abstract knowledge was acquired more quickly than specific knowledge. However, in the high dispersion condition, specific knowledge was acquired more quickly. These findings support the idea that the blessing of abstraction depends on the dispersion of specific variables. Gershman suggests that hierarchical Bayesian models provide a framework for understanding how learning occurs at different levels of abstraction simultaneously. A short recent paper that touches on how categories are arranged hierarchically. |
Questions under discussion
- What is a category? How does it relate to a concept?
- Why is categorization so central for cognition?
