Barnes, D., & Hampson, P. (1993). Stimulus equivalence and connectionism: Implications for behavior analysis and cognitive science. The Psychological Record, 43, 617-638.
The authors, researchers at the University of Cork, conducted research concerning connectionist models of stimulus equivalence phenomena. The results demonstrated that the network responded "in a class and context consistent manner" on 15 untrained tasks, following limited training on only 6 tasks (i.e., Stimulus Set 8); furthermore, it was shown that this performance required that explicit training be provided on at least three of seven stimulus sets, each of which employed the same 21 types of task as Stimulus Set 8. In general, the accuracy of its test performance on the eighth stimulus set depended on both the presence and amount of explicit training on Stimulus Sets 1 to 7. Thus, explicit training is a necessary precursor which permits the network to solve new problems that require the identification of new instances of sameness, difference, and opposition.