A Computational Approach for Predicting Individuals' Response Patterns in Human Syllogistic Reasoning

AbstractOne challenge within cognitive psychology on human reasoning is modeling a wide range of tasks within a certain theory. Recently, a meta-study on human syllogistic reasoning has shown that none of the established theories seemed to adequately match the human data. Possible reasons for this sobering result could be that (i) these theories do not account for differences among reasoners and (ii) they presuppose the same assumptions throughout all 64 syllogistic reasoning tasks. In this paper, we will address both aspects by proposing "clustering by principle patterns" for syllogistic reasoning based on cognitive principles, which have their roots in the literature of cognitive science and philosophy of language. These principles determine how the tasks are formally represented within the weak completion semantics, a logic programming approach that has already been successfully applied for modeling various human reasoning episodes. We will develop a generic cognitive characterization of (i) the reasoners and (ii) the tasks by integrating the results of a machine learning algorithm with underlying cognitive principles. These principles provide a cognitively plausible characterization of the response patterns that cover the population of reasoners. "Clustering by principle patterns" achieves the highest prediction accuracy compared to the available benchmark models, and gives insights to the differences among (i) the reasoners and among (ii) the explaining principles throughout the tasks.

Return to previous page