Improving Cognitive Models for Syllogistic Reasoning

AbstractMultiple cognitive theories make conflicting explainations for human reasoning on syllogistic problems. The evaluation and comparison of these theories can be performed by conceiving them as predictive models. Model evaluation often employs static sets of predictions rather than full implementations of the theories. However, most theories predict different responses depending on the state of their internal parameters. Disregarding the theories’ capabilities to adapt parameters to different reasoners leads to an incomplete picture of their predictive power. This article provides parameterized algorithmic formalizations and implementations of some syllogistic theories regarding the syllogistic single-response task. Evaluations reveal a substantial improvement for most cognitive theories being made adaptive over their original static predictions. The best performing implementations are PHM, mReasoner and Verbal Models, which almost reach the MFA benchmark. The results show that there exist heuristic and model-based theories which are able to capture a large portion of the patterns in syllogistic reasoning data.

Return to previous page