Postgraduates work in progress
Speaker: Nina Poth
Title: Similarities and Bayesian inference
Abstract: It is popular amongst psychological models to explain categorisation with psychological similarity. But what is similarity? There are two influential approaches to this: psychological similarity is (i) an inverse function of geometric distance in a psychological space (Shepard 1987) and (ii) a weighted function of common and distinct sets of features (Tversky 1977). (ii) argues that similarity should be constrained by the metric axioms because under this assumption, one can explain categorisation behaviour of various species across a variety of contexts. But (ii) presents data showing that people’s similarity judgements violate some of the metric axioms (e.g., people usually judge New York to be less similar to Tel Aviv than Tel Aviv to New York, thereby violating the symmetry axiom). (ii) explains this by giving the shared features in the first context less weight than in the second context.
Many models explain categorisation using either of these approaches, but not both. Can they be unified?
In this talk, I present Tenenbaum & Griffith’ (2001) Bayesian theory of generalisation and similarity, which, as they claim, unifies (i) and (ii). I challenge this claim. I show that the Bayesian theory implicitly relies on a theory of similarity and argue that the type of unification in question is (at most) empirical but not theoretical.
- Shepard, R. N. 1987. Toward a universal law of generalization for psychological science. Science, 237(4820), 1317-1323.
- Tenenbaum, J. B., & Griffiths, T. L. (2001). Generalization, similarity, and Bayesian inference. Behavioral and brain sciences, 24(4), 629-640.
- Tversky, A. (1977). Features of similarity. Psychological review, 84(4), 327.