Using Reinforcement-Based Models Of Transitive Inference To Stimulate Primate Data

Olga F. Lazareva (Mentor)The ability of reinforcement-based models to predict transitive inference behavior has been extensively documented. However, most of the research on models of TI has relied on data from pigeons. Pigeons are trained sequentially (first A+ B-, then B+ C-, etc.); the number of stimuli is limited to 5; and correction trials are used during training. In contrast, primates can be trained with 7 stimuli (A to G) introduced simultaneously with no correction trials. Would the models still accurately account for these data? We will present the results of simulations with two configural models of TI (Wynne, 1995; Siemann & Delius, 1998) used to predict transitive choices of rhesus monkeys after successive and simultaneous training in 7-item series task, as well as after a list-linking procedure. Our results suggest that reinforcement-based models do not account for symbolic distance effect after backward training (first F+ G-, then E+ F-, and so on). Surprisingly, the models provide a relatively good fit of list linking data, contrary to common belief that list-linking design presents a challenge for such models.Drake University, Department of Psychology, College of Arts & Science