TEST OF A SIMPLE-MODEL OF TRANSITIVE INFE RENCE USING A NONVERBAL FORM OF PRESENTATION

Transitive inference is the ability to derive a relationship (B > D) f rom the adjacent relations (A > B, B > C, C > D, D > E) of a stimulus series. Traditionally, such tasks have been presented verbally. Growin g interest in animal cognition, however, led to the development of a n onverbal version, where the verbal premises are conveyed through the t raining pairs A+B-, B+C-, C+D- and D+E-. Choices of the positive stimu lus are rewarded, while choices of the negative stimulus are punished. In novel test pairs, items B and D are presented together and subject s are expected to choose transitively item B over D. Using this task i t has been shown that several non-human species, as well as younger ch ildren and adults, are able to solve such problems. Conditioning model s, proposing that transitive choices are based on the graded weights t he different stimuli can acquire through learning, yield a very simple explanation. In order to test the validity of this assumption the sti mulus relations of linear series were changed in the experiments of th e present article. In the first experiment human subjects were trained with the overlapping pairs of two 4-term series that shared the same end stimuli but differed in the middle stimuli. Test decisions between items belonging to separate series could be explained by a simple con ditioning model. In the second experiment subjects were first taught t he 5 overlapping training pairs of a linear series. After transitivity tests a sixth pair was introduced, altering the linear series to a ci rcle, and a second test was carried out. Rather unexpectedly, subjects that had chosen the higher-ranking items in the first test generally continued to prefer these stimuli also in the second test. In order to explain these results a simple conditioning model was modified, predi citing the empirical observations reasonably well.