Generating Interpretable Hypotheses Based on Syllogistic Patterns