Most neurosymbolic systems assume a ready-made symbolic specification. The new work highlights that this assumption skips a difficult step: turning the needs of multiple stakeholders into formal constraints.
The authors propose an architecture that uses ontologies to mediate this acquisition process. By structuring stakeholder input through an ontology, the system can support eliciting and formalizing constraints before neurosymbolic learning begins.
The abstract does not include experimental results, so the contribution is the architectural proposal itself. It reframes an underappreciated bottleneck in neurosymbolic research.