Extension of benefits to complex logical structures

Determine whether the performance benefits of the structured decomposition framework—where large language models populate OWL 2 ABox assertions and SWRL rules provide deterministic verification—extend to tasks whose classification logic involves disjunctions, explicit negation, or nested quantifiers rather than simple conjunctive predicates.

Background

The paper evaluates the framework on three binary classification tasks whose decision rules are modeled as straightforward conjunctions of predicates encoded in SWRL. While these experiments demonstrate aggregate improvements over few-shot prompting and highlight the contribution of symbolic verification, they do not test more complex logical forms.

The authors explicitly note that their current predicates are simple conjunctions and that it is unknown whether the observed benefits carry over to richer logical structures such as disjunctions, negation, or nested quantifiers, leaving this as an unresolved question.