Roger Newland Shepard was one of the most creative cognitive scientists of the twentieth century. Working at Bell Labs and later Stanford, he combined elegant experiments with deep mathematical theorizing to reveal the geometric structure of mental representations. His universal law of generalization, his development of multidimensional scaling, and his mental rotation experiments each rank among the most important contributions to cognitive science.
The Universal Law of Generalization
g = probability of generalizing a learned response
d = distance in psychological space
c = sensitivity parameter
Shepard's 1987 Science paper demonstrated that across species, stimulus types, and paradigms, the probability of generalizing a learned response decays as a negative exponential function of psychological distance. He derived this from Bayesian reasoning about the probability that two stimuli belong to the same consequential region -- making it one of the few psychological laws with both universal empirical regularity and rational theoretical justification.
Multidimensional Scaling
d_ij = ||x_i - x_j|| (Euclidean distance)
Monotone: s_ij >= s_ik implies d_ij <= d_ik
Minimize stress = sqrt[Sum(d_ij - d_hat_ij)^2 / Sum d_ij^2]
In the early 1960s, Shepard developed nonmetric multidimensional scaling, a method for recovering geometric structure from ordinal similarity judgments. The key insight was that only rank order of similarities is needed to recover the spatial configuration. Kruskal formalized the approach with an optimization algorithm, and the resulting Shepard-Kruskal MDS became one of the most widely used methods for studying mental representations.
Shepard and Metzler's 1971 experiment demonstrated that time to judge whether two three-dimensional objects are identical increases linearly with angular difference -- as if the mind rotates a mental image at constant rate. This provided strong evidence for analog mental representations and launched mental imagery as a central topic in cognitive psychology.
Legacy and Impact
The universal law of generalization has been called the closest psychology has to a law like those in physics. Multidimensional scaling transformed how psychologists study similarity and mental representation. Shepard's concept of second-order isomorphism -- that relations among mental representations mirror relations among objects -- remains a foundational principle of representational theories of mind.