A Transfer of Explicitly and Nonexplicitly Trained Sequence Responses Through Equivalence Relations: An Experimental Demonstration and Connectionist Model

Abstract

Eight college students were trained in a sequential response to two nonsense syllable pairs B1–B2 and B2–B3. They were then tested to determine whether an untrained, novel three-response sequence emerged as a result of the spontaneous combination of the two explicitly trained two-response sequences (i.e., B1–B2 + B2–B3 becomes B1–B2–B3). All subjects demonstrated this three-response sequence performance. Four experimental subjects were then trained in a set of conditional discriminations, using a matching-to-sample procedure (i.e., A1–B1, A1–C1, A2–B2, A2–C2, A3–B3, A3–C3). All four subsequently displayed the formation of three equivalence relations in nonreinforced testing (i.e., A1–B1–C1, A2–B2–C2, A3–B3–C3). This was followed by testing for a transfer, through equivalence, of the sequential responses, for both nonsense syllable pairs and triples. All experimental subjects showed this transfer (i.e., C1–C2, C2–C3, and C1–C2-C3). The control subjects, however, who had not received the conditional discrimination training and equivalence testing, all failed to show this transfer. A connectionist model was then designed to simulate the obtained data. After suitable explicit training on up to four stimulus sets, the model was capable of correctly responding to the sequence, equivalence, and transfer tests, after exposure to only the training stages of a fifth set. The wider implications of connectionism for behavior analysis are also considered.