Figure 1.

Paradigm and possible gradients of generalization. (a) The preconditioning paradigm has three phases: (1) association learning, (2) reward learning, and (3) decision-making. In the association phase, participants are exposed to six pairs of faces in a statistical learning paradigm. This establishes associations between S1 and S2. In the pursuant reward phase, participants learn through conditioning that half of the S2 are associated with monetary reward (S2+), whereas the other S2 stimuli predict a neutral outcome (S2−). No S1 are shown at this stage. In the final decision phase, participants perform a 2-alternative forced choice (2AFC) task to decide between two stimuli (S1+ vs. S1− or S2+ vs. S2−) for a possible monetary win. In addition, we present S1 stimuli with new head orientations (S1*+ vs. S1*−). The tendency to choose S1+ over S1− indicates transfer of reward from S2 to S1, and thus memory guidance of decision making. If subjects also chose S1*+ over S1*−, this indicates automatic generalization from S1 to S1*. (b) Memory guidance of decision making by association could be specific for the images shown during the association phase, i.e., memory guidance would occur on a basis of a pictorial format of representation. Alternatively, if there is generalization to new views, this could either follow a narrow gradient of generalization to mirror-symmetric versions of the same head orientation as in the original association; or, memory guidance could rely on view-invariant associations, generalizing to all head orientations. Faces in a and b are taken from “25 White Faces Manipulated on Trustworthiness (Steps of 1 SD)” data set from the Social Perception Lab (http://tlab.princeton.edu/databases/), and are reproduced with permission.