Two tests of instance-based and abstract rule-based accounts of invariant learning

Two manipulations are argued to distinguish between instance-based and abstract rule-based accounts of invariant learning. Three experiments examined the effects of manipulating the type of invariant feature in the learning set, and the type of training schedules prior to test. In line with traditional research, selection bias at test was present when the invariant was the consistent inclusion of a stimulus item in the learning set. However, the degree of bias was identical when the invariant was the consistent exclusion of the stimulus item. In addition, negative transfer of training was observed when subjects were trained on one learning set and then shifted training to the opposite learning set, but no positive transfer of training was observed when subjects were trained on one learning set and then continued training using the same learning set. These results are argued to be evidence for instance-based accounts of invariant learning. © 1998 Elsevier Science B.V. All rights reserved.