Computational Simulation of Equivalence Class Formation Using the go/no-go Procedure with Compound Stimuli

Research about equivalence has commonly utilized human participants as experimental subjects. More recently, computational models have been capable of reproducing performances observed in experiments with humans. The computational model often utilized is called RELNET, and it simulates training and testing trials of conditional relations using the matching-to-sample procedure (MTS). The differentiation between sample stimulus and comparison stimuli, indispensable in MTS, implies operational difficulties for simulations. For this reason, new studies seek to utilize alternative procedures to MTS, which do not differentiate the functions of the antecedent stimuli. This work evaluated the possibility of developing a new computational model to simulate equivalence class formation using the go/no-go procedure with compound stimuli. In Experiment 1, artificial neural networks were utilized to simulate training of the AB and BC relations as well as the testing of the AC relation. The results showed that four out of six runs demonstrated equivalence class formation. Experiment 2 evaluated whether the additional class training performed in Experiment 1, which was analogous to the simulation of pre-experimental experience of human participants, would be essential for simulating the establishment of equivalence classes. It was found that it was not possible to simulate equivalence class formation without the additional class training. Altogether, the experiments show that it is possible to simulate equivalence class formation using the go/no-go procedure with compound stimuli and that it is necessary to conduct additional class training. The model developed is, therefore, an alternative to RELNET for the study of equivalence relations using computational simulations.